@arizeai/phoenix-evals 0.9.0 → 1.0.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +23 -23
- package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +58 -0
- package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +9 -1
- package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js +14 -3
- package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +16 -2
- package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js +15 -1
- package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/esm/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/esm/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.js +59 -0
- package/dist/esm/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -11
- package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js +44 -8
- package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/esm/__generated__/default_templates/index.d.ts +2 -0
- package/dist/esm/__generated__/default_templates/index.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/index.js +2 -0
- package/dist/esm/__generated__/default_templates/index.js.map +1 -1
- package/dist/esm/__generated__/types.d.ts +1 -1
- package/dist/esm/__generated__/types.d.ts.map +1 -1
- package/dist/esm/core/EvaluatorBase.d.ts +2 -2
- package/dist/esm/core/EvaluatorBase.d.ts.map +1 -1
- package/dist/esm/core/FunctionEvaluator.d.ts +1 -1
- package/dist/esm/core/FunctionEvaluator.d.ts.map +1 -1
- package/dist/esm/core/FunctionEvaluator.js.map +1 -1
- package/dist/esm/helpers/asEvaluatorFn.d.ts +1 -1
- package/dist/esm/helpers/asEvaluatorFn.d.ts.map +1 -1
- package/dist/esm/helpers/asEvaluatorFn.js.map +1 -1
- package/dist/esm/helpers/createEvaluator.d.ts +2 -2
- package/dist/esm/helpers/createEvaluator.d.ts.map +1 -1
- package/dist/esm/helpers/createEvaluator.js.map +1 -1
- package/dist/esm/helpers/toEvaluationResult.d.ts +1 -1
- package/dist/esm/helpers/toEvaluationResult.d.ts.map +1 -1
- package/dist/esm/llm/ClassificationEvaluator.d.ts +3 -3
- package/dist/esm/llm/ClassificationEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/ClassificationEvaluator.js.map +1 -1
- package/dist/esm/llm/LLMEvaluator.d.ts +1 -1
- package/dist/esm/llm/LLMEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createClassificationEvaluator.d.ts +1 -1
- package/dist/esm/llm/createClassificationEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createClassificationEvaluator.js.map +1 -1
- package/dist/esm/llm/createClassifierFn.d.ts +1 -1
- package/dist/esm/llm/createClassifierFn.d.ts.map +1 -1
- package/dist/esm/llm/createClassifierFn.js.map +1 -1
- package/dist/esm/llm/createConcisenessEvaluator.d.ts +43 -0
- package/dist/esm/llm/createConcisenessEvaluator.d.ts.map +1 -0
- package/dist/esm/llm/createConcisenessEvaluator.js +39 -0
- package/dist/esm/llm/createConcisenessEvaluator.js.map +1 -0
- package/dist/esm/llm/createCorrectnessEvaluator.d.ts +2 -2
- package/dist/esm/llm/createCorrectnessEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createCorrectnessEvaluator.js.map +1 -1
- package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts +2 -2
- package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createDocumentRelevanceEvaluator.js.map +1 -1
- package/dist/esm/llm/createFaithfulnessEvaluator.d.ts +2 -2
- package/dist/esm/llm/createFaithfulnessEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createFaithfulnessEvaluator.js.map +1 -1
- package/dist/esm/llm/createHallucinationEvaluator.d.ts +2 -2
- package/dist/esm/llm/createHallucinationEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createHallucinationEvaluator.js.map +1 -1
- package/dist/esm/llm/createRefusalEvaluator.d.ts +44 -0
- package/dist/esm/llm/createRefusalEvaluator.d.ts.map +1 -0
- package/dist/esm/llm/createRefusalEvaluator.js +40 -0
- package/dist/esm/llm/createRefusalEvaluator.js.map +1 -0
- package/dist/esm/llm/createToolInvocationEvaluator.d.ts +2 -2
- package/dist/esm/llm/createToolInvocationEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createToolInvocationEvaluator.js.map +1 -1
- package/dist/esm/llm/createToolResponseHandlingEvaluator.d.ts +2 -2
- package/dist/esm/llm/createToolResponseHandlingEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createToolResponseHandlingEvaluator.js.map +1 -1
- package/dist/esm/llm/createToolSelectionEvaluator.d.ts +2 -2
- package/dist/esm/llm/createToolSelectionEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createToolSelectionEvaluator.js.map +1 -1
- package/dist/esm/llm/generateClassification.d.ts +2 -2
- package/dist/esm/llm/generateClassification.d.ts.map +1 -1
- package/dist/esm/llm/generateClassification.js +1 -1
- package/dist/esm/llm/generateClassification.js.map +1 -1
- package/dist/esm/llm/index.d.ts +2 -0
- package/dist/esm/llm/index.d.ts.map +1 -1
- package/dist/esm/llm/index.js +2 -0
- package/dist/esm/llm/index.js.map +1 -1
- package/dist/esm/template/applyTemplate.d.ts +1 -1
- package/dist/esm/template/applyTemplate.d.ts.map +1 -1
- package/dist/esm/template/applyTemplate.js +1 -1
- package/dist/esm/template/applyTemplate.js.map +1 -1
- package/dist/esm/template/getTemplateVariables.d.ts +1 -1
- package/dist/esm/template/getTemplateVariables.d.ts.map +1 -1
- package/dist/esm/template/getTemplateVariables.js.map +1 -1
- package/dist/esm/tsconfig.esm.tsbuildinfo +1 -1
- package/dist/esm/types/evals.d.ts +5 -5
- package/dist/esm/types/evals.d.ts.map +1 -1
- package/dist/esm/types/otel.d.ts +1 -1
- package/dist/esm/types/otel.d.ts.map +1 -1
- package/dist/esm/utils/bindEvaluator.d.ts +2 -2
- package/dist/esm/utils/bindEvaluator.d.ts.map +1 -1
- package/dist/esm/utils/objectMappingUtils.d.ts +1 -1
- package/dist/esm/utils/objectMappingUtils.d.ts.map +1 -1
- package/dist/esm/utils/objectMappingUtils.js.map +1 -1
- package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -0
- package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +9 -1
- package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js +14 -3
- package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +16 -2
- package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js +15 -1
- package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/src/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/src/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.js +62 -0
- package/dist/src/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -11
- package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js +44 -8
- package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
- package/dist/src/__generated__/default_templates/index.d.ts +2 -0
- package/dist/src/__generated__/default_templates/index.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/index.js +5 -1
- package/dist/src/__generated__/default_templates/index.js.map +1 -1
- package/dist/src/__generated__/types.d.ts +1 -1
- package/dist/src/__generated__/types.d.ts.map +1 -1
- package/dist/src/core/EvaluatorBase.d.ts +2 -2
- package/dist/src/core/EvaluatorBase.d.ts.map +1 -1
- package/dist/src/core/FunctionEvaluator.d.ts +1 -1
- package/dist/src/core/FunctionEvaluator.d.ts.map +1 -1
- package/dist/src/core/FunctionEvaluator.js.map +1 -1
- package/dist/src/helpers/asEvaluatorFn.d.ts +1 -1
- package/dist/src/helpers/asEvaluatorFn.d.ts.map +1 -1
- package/dist/src/helpers/asEvaluatorFn.js.map +1 -1
- package/dist/src/helpers/createEvaluator.d.ts +2 -2
- package/dist/src/helpers/createEvaluator.d.ts.map +1 -1
- package/dist/src/helpers/createEvaluator.js.map +1 -1
- package/dist/src/helpers/toEvaluationResult.d.ts +1 -1
- package/dist/src/helpers/toEvaluationResult.d.ts.map +1 -1
- package/dist/src/llm/ClassificationEvaluator.d.ts +3 -3
- package/dist/src/llm/ClassificationEvaluator.d.ts.map +1 -1
- package/dist/src/llm/ClassificationEvaluator.js.map +1 -1
- package/dist/src/llm/LLMEvaluator.d.ts +1 -1
- package/dist/src/llm/LLMEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createClassificationEvaluator.d.ts +1 -1
- package/dist/src/llm/createClassificationEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createClassificationEvaluator.js.map +1 -1
- package/dist/src/llm/createClassifierFn.d.ts +1 -1
- package/dist/src/llm/createClassifierFn.d.ts.map +1 -1
- package/dist/src/llm/createClassifierFn.js.map +1 -1
- package/dist/src/llm/createConcisenessEvaluator.d.ts +43 -0
- package/dist/src/llm/createConcisenessEvaluator.d.ts.map +1 -0
- package/dist/src/llm/createConcisenessEvaluator.js +50 -0
- package/dist/src/llm/createConcisenessEvaluator.js.map +1 -0
- package/dist/src/llm/createCorrectnessEvaluator.d.ts +2 -2
- package/dist/src/llm/createCorrectnessEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createCorrectnessEvaluator.js.map +1 -1
- package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts +2 -2
- package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createDocumentRelevanceEvaluator.js.map +1 -1
- package/dist/src/llm/createFaithfulnessEvaluator.d.ts +2 -2
- package/dist/src/llm/createFaithfulnessEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createFaithfulnessEvaluator.js.map +1 -1
- package/dist/src/llm/createHallucinationEvaluator.d.ts +2 -2
- package/dist/src/llm/createHallucinationEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createHallucinationEvaluator.js.map +1 -1
- package/dist/src/llm/createRefusalEvaluator.d.ts +44 -0
- package/dist/src/llm/createRefusalEvaluator.d.ts.map +1 -0
- package/dist/src/llm/createRefusalEvaluator.js +51 -0
- package/dist/src/llm/createRefusalEvaluator.js.map +1 -0
- package/dist/src/llm/createToolInvocationEvaluator.d.ts +2 -2
- package/dist/src/llm/createToolInvocationEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createToolInvocationEvaluator.js.map +1 -1
- package/dist/src/llm/createToolResponseHandlingEvaluator.d.ts +2 -2
- package/dist/src/llm/createToolResponseHandlingEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createToolResponseHandlingEvaluator.js.map +1 -1
- package/dist/src/llm/createToolSelectionEvaluator.d.ts +2 -2
- package/dist/src/llm/createToolSelectionEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createToolSelectionEvaluator.js.map +1 -1
- package/dist/src/llm/generateClassification.d.ts +2 -2
- package/dist/src/llm/generateClassification.d.ts.map +1 -1
- package/dist/src/llm/generateClassification.js +1 -1
- package/dist/src/llm/generateClassification.js.map +1 -1
- package/dist/src/llm/index.d.ts +2 -0
- package/dist/src/llm/index.d.ts.map +1 -1
- package/dist/src/llm/index.js +2 -0
- package/dist/src/llm/index.js.map +1 -1
- package/dist/src/template/applyTemplate.d.ts +1 -1
- package/dist/src/template/applyTemplate.d.ts.map +1 -1
- package/dist/src/template/applyTemplate.js +1 -1
- package/dist/src/template/applyTemplate.js.map +1 -1
- package/dist/src/template/getTemplateVariables.d.ts +1 -1
- package/dist/src/template/getTemplateVariables.d.ts.map +1 -1
- package/dist/src/template/getTemplateVariables.js.map +1 -1
- package/dist/src/types/evals.d.ts +5 -5
- package/dist/src/types/evals.d.ts.map +1 -1
- package/dist/src/types/otel.d.ts +1 -1
- package/dist/src/types/otel.d.ts.map +1 -1
- package/dist/src/utils/bindEvaluator.d.ts +2 -2
- package/dist/src/utils/bindEvaluator.d.ts.map +1 -1
- package/dist/src/utils/objectMappingUtils.d.ts +1 -1
- package/dist/src/utils/objectMappingUtils.d.ts.map +1 -1
- package/dist/src/utils/objectMappingUtils.js.map +1 -1
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +38 -39
- package/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +60 -0
- package/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +9 -1
- package/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.ts +14 -3
- package/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +16 -2
- package/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.ts +15 -1
- package/src/__generated__/default_templates/REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.ts +61 -0
- package/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.ts +61 -11
- package/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.ts +44 -8
- package/src/__generated__/default_templates/index.ts +2 -0
- package/src/__generated__/types.ts +1 -1
- package/src/core/EvaluatorBase.ts +2 -2
- package/src/core/FunctionEvaluator.ts +5 -2
- package/src/helpers/asEvaluatorFn.ts +1 -2
- package/src/helpers/createEvaluator.ts +2 -3
- package/src/helpers/toEvaluationResult.ts +1 -1
- package/src/llm/ClassificationEvaluator.ts +4 -5
- package/src/llm/LLMEvaluator.ts +1 -1
- package/src/llm/createClassificationEvaluator.ts +1 -2
- package/src/llm/createClassifierFn.ts +1 -2
- package/src/llm/createConcisenessEvaluator.ts +71 -0
- package/src/llm/createCorrectnessEvaluator.ts +2 -3
- package/src/llm/createDocumentRelevanceEvaluator.ts +2 -3
- package/src/llm/createFaithfulnessEvaluator.ts +2 -3
- package/src/llm/createHallucinationEvaluator.ts +2 -3
- package/src/llm/createRefusalEvaluator.ts +70 -0
- package/src/llm/createToolInvocationEvaluator.ts +2 -3
- package/src/llm/createToolResponseHandlingEvaluator.ts +2 -3
- package/src/llm/createToolSelectionEvaluator.ts +2 -3
- package/src/llm/generateClassification.ts +5 -5
- package/src/llm/index.ts +2 -0
- package/src/template/applyTemplate.ts +2 -3
- package/src/template/getTemplateVariables.ts +2 -2
- package/src/types/evals.ts +5 -5
- package/src/types/otel.ts +1 -1
- package/src/utils/bindEvaluator.ts +2 -2
- package/src/utils/objectMappingUtils.ts +2 -2
|
@@ -1,7 +1,9 @@
|
|
|
1
|
+
export { CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
1
2
|
export { CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
2
3
|
export { DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG } from "./DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
3
4
|
export { FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
4
5
|
export { HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG } from "./HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
6
|
+
export { REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG } from "./REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
5
7
|
export { TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
6
8
|
export { TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
7
9
|
export { TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG";
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,kDAAkD,EAAE,MAAM,sDAAsD,CAAC;AAC1H,OAAO,EAAE,4CAA4C,EAAE,MAAM,gDAAgD,CAAC;AAC9G,OAAO,EAAE,6CAA6C,EAAE,MAAM,iDAAiD,CAAC;AAChH,OAAO,EAAE,+CAA+C,EAAE,MAAM,mDAAmD,CAAC;AACpH,OAAO,EAAE,sDAAsD,EAAE,MAAM,0DAA0D,CAAC;AAClI,OAAO,EAAE,8CAA8C,EAAE,MAAM,kDAAkD,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,kDAAkD,EAAE,MAAM,sDAAsD,CAAC;AAC1H,OAAO,EAAE,4CAA4C,EAAE,MAAM,gDAAgD,CAAC;AAC9G,OAAO,EAAE,6CAA6C,EAAE,MAAM,iDAAiD,CAAC;AAChH,OAAO,EAAE,uCAAuC,EAAE,MAAM,2CAA2C,CAAC;AACpG,OAAO,EAAE,+CAA+C,EAAE,MAAM,mDAAmD,CAAC;AACpH,OAAO,EAAE,sDAAsD,EAAE,MAAM,0DAA0D,CAAC;AAClI,OAAO,EAAE,8CAA8C,EAAE,MAAM,kDAAkD,CAAC"}
|
|
@@ -1,7 +1,9 @@
|
|
|
1
1
|
"use strict";
|
|
2
2
|
// This file is generated. Do not edit by hand.
|
|
3
3
|
Object.defineProperty(exports, "__esModule", { value: true });
|
|
4
|
-
exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG = exports.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG = void 0;
|
|
4
|
+
exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG = exports.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG = exports.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG = void 0;
|
|
5
|
+
var CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
6
|
+
Object.defineProperty(exports, "CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
5
7
|
var CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
6
8
|
Object.defineProperty(exports, "CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG_1.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
7
9
|
var DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
@@ -10,6 +12,8 @@ var FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./FAITHFULNESS_CLA
|
|
|
10
12
|
Object.defineProperty(exports, "FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG_1.FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
11
13
|
var HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
12
14
|
Object.defineProperty(exports, "HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG_1.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
15
|
+
var REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
16
|
+
Object.defineProperty(exports, "REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG_1.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
13
17
|
var TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
14
18
|
Object.defineProperty(exports, "TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG_1.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG; } });
|
|
15
19
|
var TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG");
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAE/C,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,2HAA0H;AAAjH,wLAAA,kDAAkD,OAAA;AAC3D,+GAA8G;AAArG,4KAAA,4CAA4C,OAAA;AACrD,iHAAgH;AAAvG,8KAAA,6CAA6C,OAAA;AACtD,qHAAoH;AAA3G,kLAAA,+CAA+C,OAAA;AACxD,mIAAkI;AAAzH,gMAAA,sDAAsD,OAAA;AAC/D,mHAAkH;AAAzG,gLAAA,8CAA8C,OAAA"}
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAE/C,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,2HAA0H;AAAjH,wLAAA,kDAAkD,OAAA;AAC3D,+GAA8G;AAArG,4KAAA,4CAA4C,OAAA;AACrD,iHAAgH;AAAvG,8KAAA,6CAA6C,OAAA;AACtD,qGAAoG;AAA3F,kKAAA,uCAAuC,OAAA;AAChD,qHAAoH;AAA3G,kLAAA,+CAA+C,OAAA;AACxD,mIAAkI;AAAzH,gMAAA,sDAAsD,OAAA;AAC/D,mHAAkH;AAAzG,gLAAA,8CAA8C,OAAA"}
|
|
@@ -2,7 +2,7 @@ import type { PromptTemplate } from "../types/templating";
|
|
|
2
2
|
export type ClassificationEvaluatorConfig = {
|
|
3
3
|
name: string;
|
|
4
4
|
description: string;
|
|
5
|
-
optimizationDirection: "MINIMIZE" | "MAXIMIZE";
|
|
5
|
+
optimizationDirection: "MINIMIZE" | "MAXIMIZE" | "NEUTRAL";
|
|
6
6
|
template: PromptTemplate;
|
|
7
7
|
choices: Record<string, number>;
|
|
8
8
|
};
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/__generated__/types.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAE1D,MAAM,MAAM,6BAA6B,GAAG;IAC1C,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,qBAAqB,EAAE,UAAU,GAAG,UAAU,CAAC;
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/__generated__/types.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAE1D,MAAM,MAAM,6BAA6B,GAAG;IAC1C,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,qBAAqB,EAAE,UAAU,GAAG,UAAU,GAAG,SAAS,CAAC;IAC3D,QAAQ,EAAE,cAAc,CAAC;IACzB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjC,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { CreateEvaluatorArgs, EvaluationKind, EvaluationResult, EvaluatorInterface, OptimizationDirection, TelemetryConfig } from "../types";
|
|
2
|
-
import { ObjectMapping } from "../types/data";
|
|
1
|
+
import type { CreateEvaluatorArgs, EvaluationKind, EvaluationResult, EvaluatorInterface, OptimizationDirection, TelemetryConfig } from "../types";
|
|
2
|
+
import type { ObjectMapping } from "../types/data";
|
|
3
3
|
/**
|
|
4
4
|
* Base class for all evaluators
|
|
5
5
|
*/
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"EvaluatorBase.d.ts","sourceRoot":"","sources":["../../../src/core/EvaluatorBase.ts"],"names":[],"mappings":"AAAA,OAAO,
|
|
1
|
+
{"version":3,"file":"EvaluatorBase.d.ts","sourceRoot":"","sources":["../../../src/core/EvaluatorBase.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,mBAAmB,EACnB,cAAc,EACd,gBAAgB,EAChB,kBAAkB,EAClB,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAClB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAEnD;;GAEG;AACH,8BAAsB,aAAa,CACjC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,YAAW,kBAAkB,CAAC,UAAU,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,cAAc,CAAC;IAC9B,QAAQ,CAAC,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IACvD,QAAQ,CAAC,YAAY,CAAC,EAAE,aAAa,CAAC,UAAU,CAAC,CAAC;IAClD,QAAQ,CAAC,SAAS,CAAC,EAAE,eAAe,CAAC;gBACzB,EACV,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,YAAY,EACZ,SAAS,GACV,EAAE,mBAAmB,CAAC,UAAU,CAAC;IAOlC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAElE;;OAEG;IACH,QAAQ,CAAC,gBAAgB,CACvB,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,aAAa,CAAC,UAAU,CAAC;CAC7B"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { CreateEvaluatorArgs, EvaluationResult, ObjectMapping } from "../types";
|
|
1
|
+
import type { CreateEvaluatorArgs, EvaluationResult, ObjectMapping } from "../types";
|
|
2
2
|
import { EvaluatorBase } from "./EvaluatorBase";
|
|
3
3
|
type FunctionEvaluatorArgs<RecordType extends Record<string, unknown>> = CreateEvaluatorArgs<RecordType> & {
|
|
4
4
|
evaluateFn: (args: RecordType) => Promise<EvaluationResult>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"FunctionEvaluator.d.ts","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,
|
|
1
|
+
{"version":3,"file":"FunctionEvaluator.d.ts","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,mBAAmB,EACnB,gBAAgB,EAChB,aAAa,EACd,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAEhD,KAAK,qBAAqB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IACnE,mBAAmB,CAAC,UAAU,CAAC,GAAG;IAChC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;CAC7D,CAAC;AACJ;;GAEG;AACH,qBAAa,iBAAiB,CAC5B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;IACjC,QAAQ,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;gBACzD,EAAE,UAAU,EAAE,GAAG,IAAI,EAAE,EAAE,qBAAqB,CAAC,UAAU,CAAC;IAIhE,QAAQ,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAG3D,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,iBAAiB,CAAC,UAAU,CAAC;CAGjC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"FunctionEvaluator.js","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"FunctionEvaluator.js","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;;AAKA,mDAAgD;AAMhD;;GAEG;AACH,MAAa,iBAEX,SAAQ,6BAAyB;IAEjC,YAAY,EAA0D;YAA1D,EAAE,UAAU,OAA8C,EAAzC,IAAI,cAArB,cAAuB,CAAF;QAC/B,KAAK,mBAAM,IAAI,EAAG,CAAC;QACnB,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;IAC/B,CAAC;IACD,KAAK,CAAC,QAAQ,CAAC,IAAgB;QAC7B,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IACD,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,iBAAiB,iCAAM,IAAI,KAAE,YAAY,IAAG,CAAC;IAC1D,CAAC;CACF;AAhBD,8CAgBC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"asEvaluatorFn.d.ts","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;
|
|
1
|
+
{"version":3,"file":"asEvaluatorFn.d.ts","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAInD;;GAEG;AACH,wBAAgB,aAAa,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACtE,EAAE,EAAE,KAAK,GACR,WAAW,CAAC,UAAU,CAAC,CAQzB"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"asEvaluatorFn.js","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":";;
|
|
1
|
+
{"version":3,"file":"asEvaluatorFn.js","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":";;AAOA,sCAUC;AAhBD,kDAA+C;AAC/C,6DAA0D;AAE1D;;GAEG;AACH,SAAgB,aAAa,CAC3B,EAAS;IAET,OAAO,KAAK,EAAE,GAAG,IAAI,EAAE,EAAE;QACvB,IAAI,MAAM,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACzB,IAAI,IAAA,qBAAS,EAAC,MAAM,CAAC,EAAE,CAAC;YACtB,MAAM,GAAG,MAAM,MAAM,CAAC;QACxB,CAAC;QACD,OAAO,IAAA,uCAAkB,EAAC,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC;AACJ,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { EvaluatorBase } from "../core/EvaluatorBase";
|
|
2
|
-
import { EvaluationKind, OptimizationDirection, TelemetryConfig } from "../types";
|
|
1
|
+
import type { EvaluatorBase } from "../core/EvaluatorBase";
|
|
2
|
+
import type { EvaluationKind, OptimizationDirection, TelemetryConfig } from "../types";
|
|
3
3
|
type AnyFn = (...args: any[]) => any;
|
|
4
4
|
/**
|
|
5
5
|
* Options for creating a custom evaluator using {@link createEvaluator}.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createEvaluator.d.ts","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;
|
|
1
|
+
{"version":3,"file":"createEvaluator.d.ts","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAE3D,OAAO,KAAK,EACV,cAAc,EACd,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAIlB,KAAK,KAAK,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,KAAK,GAAG,CAAC;AAMrC;;;;GAIG;AACH,MAAM,MAAM,sBAAsB,GAAG;IACnC;;;;;;;;;;OAUG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;;;;;;;;;OAYG;IACH,IAAI,CAAC,EAAE,cAAc,CAAC;IACtB;;;;;;;;;;;;;;OAcG;IACH,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IAC9C;;;;;;;;;;;;;;;OAeG;IACH,SAAS,CAAC,EAAE,eAAe,CAAC;CAC7B,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,wBAAgB,eAAe,CAC7B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACpE,EAAE,SAAS,KAAK,GAAG,KAAK,EACxB,EAAE,EAAE,EAAE,EAAE,OAAO,CAAC,EAAE,sBAAsB,GAAG,aAAa,CAAC,UAAU,CAAC,CAyBrE"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createEvaluator.js","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":";;
|
|
1
|
+
{"version":3,"file":"createEvaluator.js","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":";;AA0JA,0CA4BC;AAtLD,oEAAuD;AAGvD,iEAA8D;AAM9D,mDAAgD;AAKhD,SAAS,kBAAkB;IACzB,OAAO,aAAa,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;AACpE,CAAC;AAqED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,SAAgB,eAAe,CAG7B,EAAM,EAAE,OAAgC;IACxC,MAAM,EACJ,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,SAAS,GAAG,EAAE,SAAS,EAAE,IAAI,EAAE,GAChC,GAAG,OAAO,IAAI,EAAE,CAAC;IAClB,MAAM,aAAa,GAAG,IAAI,IAAI,EAAE,CAAC,IAAI,IAAI,kBAAkB,EAAE,CAAC;IAC9D,IAAI,UAAU,GAAG,IAAA,6BAAa,EAAa,EAAE,CAAC,CAAC;IAE/C,0DAA0D;IAC1D,IAAI,SAAS,IAAI,SAAS,CAAC,SAAS,EAAE,CAAC;QACrC,UAAU,GAAG,IAAA,6BAAQ,EAAC,UAAU,EAAE;YAChC,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,IAAI,EAAE,aAAa;YACnB,IAAI,EAAE,WAAW;SAClB,CAAC,CAAC;IACL,CAAC;IACD,OAAO,IAAI,qCAAiB,CAAa;QACvC,UAAU;QACV,IAAI,EAAE,aAAa;QACnB,IAAI,EAAE,IAAI,IAAI,MAAM;QACpB,qBAAqB,EAAE,qBAAqB,IAAI,UAAU;QAC1D,SAAS;KACV,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"toEvaluationResult.d.ts","sourceRoot":"","sources":["../../../src/helpers/toEvaluationResult.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,MAAM,UAAU,CAAC;
|
|
1
|
+
{"version":3,"file":"toEvaluationResult.d.ts","sourceRoot":"","sources":["../../../src/helpers/toEvaluationResult.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,UAAU,CAAC;AAuDjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+DG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,OAAO,GAAG,gBAAgB,CAyBpE"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import {
|
|
1
|
+
import type { LanguageModel } from "ai";
|
|
2
|
+
import type { ClassificationChoicesMap, CreateClassificationEvaluatorArgs, EvaluatorFn, PromptTemplate, WithPromptTemplate } from "../types";
|
|
3
|
+
import type { ObjectMapping } from "../types/data";
|
|
3
4
|
import { LLMEvaluator } from "./LLMEvaluator";
|
|
4
|
-
import { LanguageModel } from "ai";
|
|
5
5
|
/**
|
|
6
6
|
* An LLM evaluator that performs evaluation via classification
|
|
7
7
|
*/
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"ClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAGxC,OAAO,KAAK,EACV,wBAAwB,EACxB,iCAAiC,EACjC,WAAW,EACX,cAAc,EACd,kBAAkB,EACnB,MAAM,UAAU,CAAC;AAClB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAGnD,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C;;GAEG;AACH,qBAAa,uBAAuB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC7E,SAAQ,YAAY,CAAC,UAAU,CAC/B,YAAW,kBAAkB;IAE7B,QAAQ,CAAC,WAAW,EAAE,WAAW,CAAC,UAAU,CAAC,CAAC;IAC9C,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC;IACxC;;OAEG;IACH,OAAO,CAAC,wBAAwB,CAAuB;IACvD;;OAEG;IACH,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC;IAC9B;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,wBAAwB,CAAC;gBAE/B,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC;IAS/D,QAAQ,GAAI,SAAS,UAAU,kDAM7B;IACF;;OAEG;IACH,IAAI,uBAAuB,IAAI,MAAM,EAAE,CAStC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,uBAAuB,CAAC,UAAU,CAAC;CAMvC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":";;;
|
|
1
|
+
{"version":3,"file":"ClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":";;;AAEA,0CAAmD;AASnD,oEAA0D;AAC1D,6DAA0D;AAC1D,iDAA8C;AAE9C;;GAEG;AACH,MAAa,uBACX,SAAQ,2BAAwB;IAkBhC,YAAY,IAAmD;QAC7D,KAAK,CAAC,IAAI,CAAC,CAAC;QAQd,aAAQ,GAAG,CAAC,OAAmB,EAAE,EAAE;YACjC,OAAO,IAAI,CAAC,WAAW,CACrB,IAAI,CAAC,YAAY;gBACf,CAAC,CAAC,IAAA,gCAAW,EAAa,OAAO,EAAE,IAAI,CAAC,YAAY,CAAC;gBACrD,CAAC,CAAC,OAAO,CACZ,CAAC;QACJ,CAAC,CAAC;QAbA,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,cAAc,CAAC;QAC1C,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACxB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;QAC5B,IAAI,CAAC,WAAW,GAAG,IAAA,uCAAkB,oBAChC,IAAI,EACP,CAAC;IACL,CAAC;IAQD;;OAEG;IACH,IAAI,uBAAuB;QACzB,0DAA0D;QAC1D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,wBAAwB,CAAC,EAAE,CAAC;YAClD,IAAI,CAAC,wBAAwB,GAAG,IAAA,+BAAoB,EAAC;gBACnD,QAAQ,EAAE,IAAI,CAAC,cAAc;aAC9B,CAAC,CAAC;QACL,CAAC;QACD,+BAA+B;QAC/B,OAAO,CAAC,GAAG,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC5C,CAAC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,uBAAuB,iCAC7B,IAAI,KACP,YAAY,IACZ,CAAC;IACL,CAAC;CACF;AA3DD,0DA2DC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"LLMEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/LLMEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AACtD,OAAO,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;
|
|
1
|
+
{"version":3,"file":"LLMEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/LLMEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AACtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAEvD;;GAEG;AACH,8BAAsB,YAAY,CAChC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;gBACrB,EAAE,GAAG,IAAI,EAAE,EAAE,sBAAsB,CAAC,UAAU,CAAC;CAG5D"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
2
|
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
3
|
export declare function createClassificationEvaluator<RecordType extends Record<string, unknown>>(args: CreateClassificationEvaluatorArgs<RecordType>): ClassificationEvaluator<RecordType>;
|
|
4
4
|
//# sourceMappingURL=createClassificationEvaluator.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAEpE,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAE1C,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC,GAClD,uBAAuB,CAAC,UAAU,CAAC,CAErC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":";;
|
|
1
|
+
{"version":3,"file":"createClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":";;AAGA,sEAMC;AARD,uEAAoE;AAEpE,SAAgB,6BAA6B,CAG3C,IAAmD;IAEnD,OAAO,IAAI,iDAAuB,CAAa,IAAI,CAAC,CAAC;AACvD,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createClassifierFn.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":"AACA,OAAO,
|
|
1
|
+
{"version":3,"file":"createClassifierFn.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEV,oBAAoB,EAEpB,WAAW,EACZ,MAAM,gBAAgB,CAAC;AAiBxB;;GAEG;AACH,wBAAgB,kBAAkB,CAChC,gBAAgB,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAChD,IAAI,EAAE,oBAAoB,GAAG,WAAW,CAAC,gBAAgB,CAAC,CA4B3D"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createClassifierFn.js","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":";;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"createClassifierFn.js","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":";;;;;;;;;;;;;AA0BA,gDA8BC;AAxDD,0CAA6C;AAO7C,qEAAkE;AAElE;;;GAGG;AACH,SAAS,eAAe,CACtB,OAAiC;IAEjC,MAAM,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACzC,CAAC;IACD,OAAO,MAA+B,CAAC;AACzC,CAAC;AAED;;GAEG;AACH,SAAgB,kBAAkB,CAEhC,IAA0B;IAC1B,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,cAAc,KAAc,IAAI,EAAb,IAAI,UAAK,IAAI,EAAlD,sCAA2C,CAAO,CAAC;IAEzD,OAAO,KAAK,EAAE,IAAsB,EAA6B,EAAE;QACjE,MAAM,iBAAiB,qBAClB,IAAI,CACR,CAAC;QAEF,MAAM,MAAM,GAAG,IAAA,yBAAc,EAAC;YAC5B,QAAQ,EAAE,cAAc;YACxB,SAAS,EAAE,iBAAiB;SAC7B,CAAC,CAAC;QAEH,MAAM,cAAc,GAAG,MAAM,IAAA,+CAAsB,kBACjD,KAAK,EACL,MAAM,EAAE,eAAe,CAAC,OAAO,CAAC,EAChC,MAAM,IACH,IAAI,EACP,CAAC;QAEH,mEAAmE;QACnE,MAAM,KAAK,GAAG,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QAE5C,uBACE,KAAK,IACF,cAAc,EACjB;IACJ,CAAC,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
|
+
export interface ConcisenessEvaluatorArgs<RecordType extends Record<string, unknown> = ConcisenessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
|
+
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
|
+
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
6
|
+
choices?: CreateClassificationEvaluatorArgs<RecordType>["choices"];
|
|
7
|
+
promptTemplate?: CreateClassificationEvaluatorArgs<RecordType>["promptTemplate"];
|
|
8
|
+
}
|
|
9
|
+
/**
|
|
10
|
+
* A record to be evaluated by the conciseness evaluator.
|
|
11
|
+
*/
|
|
12
|
+
export type ConcisenessEvaluationRecord = {
|
|
13
|
+
input: string;
|
|
14
|
+
output: string;
|
|
15
|
+
};
|
|
16
|
+
/**
|
|
17
|
+
* Creates a conciseness evaluator function.
|
|
18
|
+
*
|
|
19
|
+
* This function returns an evaluator that determines whether a given output
|
|
20
|
+
* is concise and free of unnecessary content such as pleasantries, hedging,
|
|
21
|
+
* meta-commentary, or redundant information.
|
|
22
|
+
*
|
|
23
|
+
* @param args - The arguments for creating the conciseness evaluator.
|
|
24
|
+
* @param args.model - The model to use for classification.
|
|
25
|
+
* @param args.choices - The possible classification choices (defaults to CONCISENESS_CHOICES).
|
|
26
|
+
* @param args.promptTemplate - The prompt template to use (defaults to CONCISENESS_TEMPLATE).
|
|
27
|
+
* @param args.telemetry - The telemetry to use for the evaluator.
|
|
28
|
+
*
|
|
29
|
+
* @returns An evaluator function that takes a {@link ConcisenessEvaluationRecord} and returns a classification result
|
|
30
|
+
* indicating whether the output is concise or verbose.
|
|
31
|
+
*
|
|
32
|
+
* @example
|
|
33
|
+
* ```ts
|
|
34
|
+
* const evaluator = createConcisenessEvaluator({ model: openai("gpt-4o-mini") });
|
|
35
|
+
* const result = await evaluator.evaluate({
|
|
36
|
+
* input: "What is the capital of France?",
|
|
37
|
+
* output: "Paris.",
|
|
38
|
+
* });
|
|
39
|
+
* console.log(result.label); // "concise" or "verbose"
|
|
40
|
+
* ```
|
|
41
|
+
*/
|
|
42
|
+
export declare function createConcisenessEvaluator<RecordType extends Record<string, unknown> = ConcisenessEvaluationRecord>(args: ConcisenessEvaluatorArgs<RecordType>): ClassificationEvaluator<RecordType>;
|
|
43
|
+
//# sourceMappingURL=createConcisenessEvaluator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"createConcisenessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createConcisenessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,wBAAwB,CACvC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,CACxE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,0BAA0B,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,EAExE,IAAI,EAAE,wBAAwB,CAAC,UAAU,CAAC,GACzC,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
|
|
@@ -0,0 +1,50 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __rest = (this && this.__rest) || function (s, e) {
|
|
3
|
+
var t = {};
|
|
4
|
+
for (var p in s) if (Object.prototype.hasOwnProperty.call(s, p) && e.indexOf(p) < 0)
|
|
5
|
+
t[p] = s[p];
|
|
6
|
+
if (s != null && typeof Object.getOwnPropertySymbols === "function")
|
|
7
|
+
for (var i = 0, p = Object.getOwnPropertySymbols(s); i < p.length; i++) {
|
|
8
|
+
if (e.indexOf(p[i]) < 0 && Object.prototype.propertyIsEnumerable.call(s, p[i]))
|
|
9
|
+
t[p[i]] = s[p[i]];
|
|
10
|
+
}
|
|
11
|
+
return t;
|
|
12
|
+
};
|
|
13
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
14
|
+
exports.createConcisenessEvaluator = createConcisenessEvaluator;
|
|
15
|
+
const default_templates_1 = require("../__generated__/default_templates");
|
|
16
|
+
const createClassificationEvaluator_1 = require("./createClassificationEvaluator");
|
|
17
|
+
/**
|
|
18
|
+
* Creates a conciseness evaluator function.
|
|
19
|
+
*
|
|
20
|
+
* This function returns an evaluator that determines whether a given output
|
|
21
|
+
* is concise and free of unnecessary content such as pleasantries, hedging,
|
|
22
|
+
* meta-commentary, or redundant information.
|
|
23
|
+
*
|
|
24
|
+
* @param args - The arguments for creating the conciseness evaluator.
|
|
25
|
+
* @param args.model - The model to use for classification.
|
|
26
|
+
* @param args.choices - The possible classification choices (defaults to CONCISENESS_CHOICES).
|
|
27
|
+
* @param args.promptTemplate - The prompt template to use (defaults to CONCISENESS_TEMPLATE).
|
|
28
|
+
* @param args.telemetry - The telemetry to use for the evaluator.
|
|
29
|
+
*
|
|
30
|
+
* @returns An evaluator function that takes a {@link ConcisenessEvaluationRecord} and returns a classification result
|
|
31
|
+
* indicating whether the output is concise or verbose.
|
|
32
|
+
*
|
|
33
|
+
* @example
|
|
34
|
+
* ```ts
|
|
35
|
+
* const evaluator = createConcisenessEvaluator({ model: openai("gpt-4o-mini") });
|
|
36
|
+
* const result = await evaluator.evaluate({
|
|
37
|
+
* input: "What is the capital of France?",
|
|
38
|
+
* output: "Paris.",
|
|
39
|
+
* });
|
|
40
|
+
* console.log(result.label); // "concise" or "verbose"
|
|
41
|
+
* ```
|
|
42
|
+
*/
|
|
43
|
+
function createConcisenessEvaluator(args) {
|
|
44
|
+
const { choices = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.choices, promptTemplate = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.template, optimizationDirection = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection, name = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.name } = args, rest = __rest(args, ["choices", "promptTemplate", "optimizationDirection", "name"]);
|
|
45
|
+
return (0, createClassificationEvaluator_1.createClassificationEvaluator)(Object.assign(Object.assign({}, rest), { promptTemplate,
|
|
46
|
+
choices,
|
|
47
|
+
optimizationDirection,
|
|
48
|
+
name }));
|
|
49
|
+
}
|
|
50
|
+
//# sourceMappingURL=createConcisenessEvaluator.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"createConcisenessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createConcisenessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAmDA,gEAmBC;AAtED,0EAAiG;AAGjG,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAgB,0BAA0B,CAGxC,IAA0C;IAE1C,MAAM,EACJ,OAAO,GAAG,+DAA2C,CAAC,OAAO,EAC7D,cAAc,GAAG,+DAA2C,CAAC,QAAQ,EACrE,qBAAqB,GAAG,+DAA2C,CAAC,qBAAqB,EACzF,IAAI,GAAG,+DAA2C,CAAC,IAAI,KAErD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
-
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
3
|
export interface CorrectnessEvaluatorArgs<RecordType extends Record<string, unknown> = CorrectnessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
4
|
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
5
|
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createCorrectnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createCorrectnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,wBAAwB,CACvC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,CACxE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,wBAAgB,0BAA0B,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,EAExE,IAAI,EAAE,wBAAwB,CAAC,UAAU,CAAC,GACzC,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createCorrectnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"createCorrectnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAkDA,gEAmBC;AArED,0EAAiG;AAGjG,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,SAAgB,0BAA0B,CAGxC,IAA0C;IAE1C,MAAM,EACJ,OAAO,GAAG,+DAA2C,CAAC,OAAO,EAC7D,cAAc,GAAG,+DAA2C,CAAC,QAAQ,EACrE,qBAAqB,GAAG,+DAA2C,CAAC,qBAAqB,EACzF,IAAI,GAAG,+DAA2C,CAAC,IAAI,KAErD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
-
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
3
|
export interface DocumentRelevanceEvaluatorArgs<RecordType extends Record<string, unknown> = DocumentRelevanceEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
4
|
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
5
|
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createDocumentRelevanceEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createDocumentRelevanceEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,8BAA8B,CAC7C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,CACnC,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,WAAW,iCAAiC;IAChD,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,gCAAgC,CAC9C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,EAEnC,IAAI,EAAE,8BAA8B,CAAC,UAAU,CAAC,GAC/C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createDocumentRelevanceEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"createDocumentRelevanceEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAqDA,4EAoBC;AAzED,0EAAwG;AAGxG,mFAAgF;AAwBhF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAgB,gCAAgC,CAI9C,IAAgD;IAEhD,MAAM,EACJ,OAAO,GAAG,sEAAkD,CAAC,OAAO,EACpE,cAAc,GAAG,sEAAkD,CAAC,QAAQ,EAC5E,qBAAqB,GAAG,sEAAkD,CAAC,qBAAqB,EAChG,IAAI,GAAG,sEAAkD,CAAC,IAAI,KAE5D,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
-
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
3
|
export interface FaithfulnessEvaluatorArgs<RecordType extends Record<string, unknown> = FaithfulnessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
4
|
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
5
|
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createFaithfulnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createFaithfulnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,yBAAyB,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,CACzE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,4BAA4B,GAAG;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AACF;;;;;GAKG;AACH,wBAAgB,2BAA2B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,EAEzE,IAAI,EAAE,yBAAyB,CAAC,UAAU,CAAC,GAC1C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createFaithfulnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"createFaithfulnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AA+BA,kEAmBC;AAlDD,0EAAkG;AAGlG,mFAAgF;AAsBhF;;;;;GAKG;AACH,SAAgB,2BAA2B,CAGzC,IAA2C;IAE3C,MAAM,EACJ,OAAO,GAAG,gEAA4C,CAAC,OAAO,EAC9D,cAAc,GAAG,gEAA4C,CAAC,QAAQ,EACtE,qBAAqB,GAAG,gEAA4C,CAAC,qBAAqB,EAC1F,IAAI,GAAG,gEAA4C,CAAC,IAAI,KAEtD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -4,8 +4,8 @@
|
|
|
4
4
|
* - 'faithful'/'unfaithful' labels instead of 'factual'/'hallucinated'
|
|
5
5
|
* - Maximizes score (1.0=faithful) instead of minimizing it
|
|
6
6
|
*/
|
|
7
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
8
|
-
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
7
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
8
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
9
9
|
export interface HallucinationEvaluatorArgs<RecordType extends Record<string, unknown> = HallucinationEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
10
10
|
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
11
11
|
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createHallucinationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createHallucinationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,0BAA0B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,CAC1E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,6BAA6B,GAAG;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAgB,4BAA4B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,EAE1E,IAAI,EAAE,0BAA0B,CAAC,UAAU,CAAC,GAC3C,uBAAuB,CAAC,UAAU,CAAC,CAuBrC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createHallucinationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"createHallucinationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;;;;;;;;;;;AAwCH,oEA2BC;AAjED,0EAAmG;AAGnG,mFAAgF;AAuBhF;;;;;;;;;;;GAWG;AACH,SAAgB,4BAA4B,CAG1C,IAA4C;IAE5C,sCAAsC;IACtC,OAAO,CAAC,IAAI,CACV,sFAAsF;QACpF,yEAAyE;QACzE,+EAA+E;QAC/E,8BAA8B,CACjC,CAAC;IAEF,MAAM,EACJ,OAAO,GAAG,iEAA6C,CAAC,OAAO,EAC/D,cAAc,GAAG,iEAA6C,CAAC,QAAQ,EACvE,qBAAqB,GAAG,iEAA6C,CAAC,qBAAqB,EAC3F,IAAI,GAAG,iEAA6C,CAAC,IAAI,KAEvD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -0,0 +1,44 @@
|
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
|
+
export interface RefusalEvaluatorArgs<RecordType extends Record<string, unknown> = RefusalEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
|
+
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
|
+
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
6
|
+
choices?: CreateClassificationEvaluatorArgs<RecordType>["choices"];
|
|
7
|
+
promptTemplate?: CreateClassificationEvaluatorArgs<RecordType>["promptTemplate"];
|
|
8
|
+
}
|
|
9
|
+
/**
|
|
10
|
+
* A record to be evaluated by the refusal evaluator.
|
|
11
|
+
*/
|
|
12
|
+
export type RefusalEvaluationRecord = {
|
|
13
|
+
input: string;
|
|
14
|
+
output: string;
|
|
15
|
+
};
|
|
16
|
+
/**
|
|
17
|
+
* Creates a refusal evaluator function.
|
|
18
|
+
*
|
|
19
|
+
* This function returns an evaluator that detects when an LLM refuses,
|
|
20
|
+
* declines, or avoids answering a user query. It is use-case agnostic:
|
|
21
|
+
* it only detects whether a refusal occurred, not whether the refusal
|
|
22
|
+
* was appropriate.
|
|
23
|
+
*
|
|
24
|
+
* @param args - The arguments for creating the refusal evaluator.
|
|
25
|
+
* @param args.model - The model to use for classification.
|
|
26
|
+
* @param args.choices - The possible classification choices (defaults to REFUSAL_CHOICES).
|
|
27
|
+
* @param args.promptTemplate - The prompt template to use (defaults to REFUSAL_TEMPLATE).
|
|
28
|
+
* @param args.telemetry - The telemetry to use for the evaluator.
|
|
29
|
+
*
|
|
30
|
+
* @returns An evaluator function that takes a {@link RefusalEvaluationRecord} and returns a classification result
|
|
31
|
+
* indicating whether the output is a refusal or an answer.
|
|
32
|
+
*
|
|
33
|
+
* @example
|
|
34
|
+
* ```ts
|
|
35
|
+
* const evaluator = createRefusalEvaluator({ model: openai("gpt-4o-mini") });
|
|
36
|
+
* const result = await evaluator.evaluate({
|
|
37
|
+
* input: "What is the capital of France?",
|
|
38
|
+
* output: "I'm sorry, I can only help with technical questions.",
|
|
39
|
+
* });
|
|
40
|
+
* console.log(result.label); // "refused" or "answered"
|
|
41
|
+
* ```
|
|
42
|
+
*/
|
|
43
|
+
export declare function createRefusalEvaluator<RecordType extends Record<string, unknown> = RefusalEvaluationRecord>(args: RefusalEvaluatorArgs<RecordType>): ClassificationEvaluator<RecordType>;
|
|
44
|
+
//# sourceMappingURL=createRefusalEvaluator.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"createRefusalEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createRefusalEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,oBAAoB,CACnC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,uBAAuB,CACpE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,uBAAuB,GAAG;IACpC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,wBAAgB,sBAAsB,CACpC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,uBAAuB,EACpE,IAAI,EAAE,oBAAoB,CAAC,UAAU,CAAC,GAAG,uBAAuB,CAAC,UAAU,CAAC,CAe7E"}
|
|
@@ -0,0 +1,51 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __rest = (this && this.__rest) || function (s, e) {
|
|
3
|
+
var t = {};
|
|
4
|
+
for (var p in s) if (Object.prototype.hasOwnProperty.call(s, p) && e.indexOf(p) < 0)
|
|
5
|
+
t[p] = s[p];
|
|
6
|
+
if (s != null && typeof Object.getOwnPropertySymbols === "function")
|
|
7
|
+
for (var i = 0, p = Object.getOwnPropertySymbols(s); i < p.length; i++) {
|
|
8
|
+
if (e.indexOf(p[i]) < 0 && Object.prototype.propertyIsEnumerable.call(s, p[i]))
|
|
9
|
+
t[p[i]] = s[p[i]];
|
|
10
|
+
}
|
|
11
|
+
return t;
|
|
12
|
+
};
|
|
13
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
14
|
+
exports.createRefusalEvaluator = createRefusalEvaluator;
|
|
15
|
+
const default_templates_1 = require("../__generated__/default_templates");
|
|
16
|
+
const createClassificationEvaluator_1 = require("./createClassificationEvaluator");
|
|
17
|
+
/**
|
|
18
|
+
* Creates a refusal evaluator function.
|
|
19
|
+
*
|
|
20
|
+
* This function returns an evaluator that detects when an LLM refuses,
|
|
21
|
+
* declines, or avoids answering a user query. It is use-case agnostic:
|
|
22
|
+
* it only detects whether a refusal occurred, not whether the refusal
|
|
23
|
+
* was appropriate.
|
|
24
|
+
*
|
|
25
|
+
* @param args - The arguments for creating the refusal evaluator.
|
|
26
|
+
* @param args.model - The model to use for classification.
|
|
27
|
+
* @param args.choices - The possible classification choices (defaults to REFUSAL_CHOICES).
|
|
28
|
+
* @param args.promptTemplate - The prompt template to use (defaults to REFUSAL_TEMPLATE).
|
|
29
|
+
* @param args.telemetry - The telemetry to use for the evaluator.
|
|
30
|
+
*
|
|
31
|
+
* @returns An evaluator function that takes a {@link RefusalEvaluationRecord} and returns a classification result
|
|
32
|
+
* indicating whether the output is a refusal or an answer.
|
|
33
|
+
*
|
|
34
|
+
* @example
|
|
35
|
+
* ```ts
|
|
36
|
+
* const evaluator = createRefusalEvaluator({ model: openai("gpt-4o-mini") });
|
|
37
|
+
* const result = await evaluator.evaluate({
|
|
38
|
+
* input: "What is the capital of France?",
|
|
39
|
+
* output: "I'm sorry, I can only help with technical questions.",
|
|
40
|
+
* });
|
|
41
|
+
* console.log(result.label); // "refused" or "answered"
|
|
42
|
+
* ```
|
|
43
|
+
*/
|
|
44
|
+
function createRefusalEvaluator(args) {
|
|
45
|
+
const { choices = default_templates_1.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.choices, promptTemplate = default_templates_1.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.template, optimizationDirection = default_templates_1.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection, name = default_templates_1.REFUSAL_CLASSIFICATION_EVALUATOR_CONFIG.name } = args, rest = __rest(args, ["choices", "promptTemplate", "optimizationDirection", "name"]);
|
|
46
|
+
return (0, createClassificationEvaluator_1.createClassificationEvaluator)(Object.assign(Object.assign({}, rest), { promptTemplate,
|
|
47
|
+
choices,
|
|
48
|
+
optimizationDirection,
|
|
49
|
+
name }));
|
|
50
|
+
}
|
|
51
|
+
//# sourceMappingURL=createRefusalEvaluator.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"createRefusalEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createRefusalEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAoDA,wDAiBC;AArED,0EAA6F;AAG7F,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AACH,SAAgB,sBAAsB,CAEpC,IAAsC;IACtC,MAAM,EACJ,OAAO,GAAG,2DAAuC,CAAC,OAAO,EACzD,cAAc,GAAG,2DAAuC,CAAC,QAAQ,EACjE,qBAAqB,GAAG,2DAAuC,CAAC,qBAAqB,EACrF,IAAI,GAAG,2DAAuC,CAAC,IAAI,KAEjD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
-
import { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
1
|
+
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
|
|
2
|
+
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
|
|
3
3
|
export interface ToolInvocationEvaluatorArgs<RecordType extends Record<string, unknown> = ToolInvocationEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
|
|
4
4
|
optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
|
|
5
5
|
name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"createToolInvocationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"createToolInvocationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,2BAA2B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,CAC3E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,8BAA8B,GAAG;IAC3C;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;OAEG;IACH,aAAa,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AACH,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,EAE3E,IAAI,EAAE,2BAA2B,CAAC,UAAU,CAAC,GAC5C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
|