judgeval 0.2.9 → 0.3.0
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/dist/constants.d.ts +2 -0
- package/dist/constants.d.ts.map +1 -0
- package/dist/constants.js +2 -0
- package/dist/constants.js.map +1 -0
- package/dist/data/example-params.d.ts +13 -0
- package/dist/data/example-params.d.ts.map +1 -0
- package/dist/data/example-params.js +12 -0
- package/dist/data/example-params.js.map +1 -0
- package/dist/data/example.d.ts +4 -0
- package/dist/data/example.d.ts.map +1 -0
- package/dist/data/example.js +10 -0
- package/dist/data/example.js.map +1 -0
- package/dist/data/index.d.ts +3 -0
- package/dist/data/index.d.ts.map +1 -0
- package/dist/data/index.js +3 -0
- package/dist/data/index.js.map +1 -0
- package/dist/env.d.ts +6 -0
- package/dist/env.d.ts.map +1 -0
- package/dist/env.js +13 -0
- package/dist/env.js.map +1 -0
- package/dist/index.d.ts +9 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +2 -0
- package/dist/index.js.map +1 -0
- package/dist/index.mjs +2 -0
- package/dist/index.mjs.map +1 -0
- package/dist/index.umd.js +2 -0
- package/dist/index.umd.js.map +1 -0
- package/dist/internal/api/index.d.ts +23 -0
- package/dist/internal/api/index.d.ts.map +1 -0
- package/dist/internal/api/index.js +132 -0
- package/dist/internal/api/index.js.map +1 -0
- package/dist/internal/api/models/BaseScorer.d.ts +23 -0
- package/dist/internal/api/models/BaseScorer.d.ts.map +1 -0
- package/dist/internal/api/models/BaseScorer.js +6 -0
- package/dist/internal/api/models/BaseScorer.js.map +1 -0
- package/dist/internal/api/models/EvalResults.d.ts +12 -0
- package/dist/internal/api/models/EvalResults.d.ts.map +1 -0
- package/dist/internal/api/models/EvalResults.js +6 -0
- package/dist/internal/api/models/EvalResults.js.map +1 -0
- package/dist/internal/api/models/EvalResultsFetch.d.ts +9 -0
- package/dist/internal/api/models/EvalResultsFetch.d.ts.map +1 -0
- package/dist/internal/api/models/EvalResultsFetch.js +6 -0
- package/dist/internal/api/models/EvalResultsFetch.js.map +1 -0
- package/dist/internal/api/models/Example.d.ts +10 -0
- package/dist/internal/api/models/Example.d.ts.map +1 -0
- package/dist/internal/api/models/Example.js +6 -0
- package/dist/internal/api/models/Example.js.map +1 -0
- package/dist/internal/api/models/ExampleEvaluationRun.d.ts +20 -0
- package/dist/internal/api/models/ExampleEvaluationRun.d.ts.map +1 -0
- package/dist/internal/api/models/ExampleEvaluationRun.js +6 -0
- package/dist/internal/api/models/ExampleEvaluationRun.js.map +1 -0
- package/dist/internal/api/models/FetchPromptScorersRequest.d.ts +8 -0
- package/dist/internal/api/models/FetchPromptScorersRequest.d.ts.map +1 -0
- package/dist/internal/api/models/FetchPromptScorersRequest.js +6 -0
- package/dist/internal/api/models/FetchPromptScorersRequest.js.map +1 -0
- package/dist/internal/api/models/FetchPromptScorersResponse.d.ts +9 -0
- package/dist/internal/api/models/FetchPromptScorersResponse.d.ts.map +1 -0
- package/dist/internal/api/models/FetchPromptScorersResponse.js +6 -0
- package/dist/internal/api/models/FetchPromptScorersResponse.js.map +1 -0
- package/dist/internal/api/models/OtelTraceSpan.d.ts +33 -0
- package/dist/internal/api/models/OtelTraceSpan.d.ts.map +1 -0
- package/dist/internal/api/models/OtelTraceSpan.js +6 -0
- package/dist/internal/api/models/OtelTraceSpan.js.map +1 -0
- package/dist/internal/api/models/PromptScorer.d.ts +14 -0
- package/dist/internal/api/models/PromptScorer.d.ts.map +1 -0
- package/dist/internal/api/models/PromptScorer.js +6 -0
- package/dist/internal/api/models/PromptScorer.js.map +1 -0
- package/dist/internal/api/models/ResolveProjectNameRequest.d.ts +8 -0
- package/dist/internal/api/models/ResolveProjectNameRequest.d.ts.map +1 -0
- package/dist/internal/api/models/ResolveProjectNameRequest.js +6 -0
- package/dist/internal/api/models/ResolveProjectNameRequest.js.map +1 -0
- package/dist/internal/api/models/ResolveProjectNameResponse.d.ts +8 -0
- package/dist/internal/api/models/ResolveProjectNameResponse.d.ts.map +1 -0
- package/dist/internal/api/models/ResolveProjectNameResponse.js +6 -0
- package/dist/internal/api/models/ResolveProjectNameResponse.js.map +1 -0
- package/dist/internal/api/models/SavePromptScorerRequest.d.ts +12 -0
- package/dist/internal/api/models/SavePromptScorerRequest.d.ts.map +1 -0
- package/dist/internal/api/models/SavePromptScorerRequest.js +6 -0
- package/dist/internal/api/models/SavePromptScorerRequest.js.map +1 -0
- package/dist/internal/api/models/SavePromptScorerResponse.d.ts +9 -0
- package/dist/internal/api/models/SavePromptScorerResponse.d.ts.map +1 -0
- package/dist/internal/api/models/SavePromptScorerResponse.js +6 -0
- package/dist/internal/api/models/SavePromptScorerResponse.js.map +1 -0
- package/dist/internal/api/models/ScorerConfig.d.ts +13 -0
- package/dist/internal/api/models/ScorerConfig.d.ts.map +1 -0
- package/dist/internal/api/models/ScorerConfig.js +6 -0
- package/dist/internal/api/models/ScorerConfig.js.map +1 -0
- package/dist/internal/api/models/ScorerData.d.ts +17 -0
- package/dist/internal/api/models/ScorerData.d.ts.map +1 -0
- package/dist/internal/api/models/ScorerData.js +6 -0
- package/dist/internal/api/models/ScorerData.js.map +1 -0
- package/dist/internal/api/models/ScorerExistsRequest.d.ts +8 -0
- package/dist/internal/api/models/ScorerExistsRequest.d.ts.map +1 -0
- package/dist/internal/api/models/ScorerExistsRequest.js +6 -0
- package/dist/internal/api/models/ScorerExistsRequest.js.map +1 -0
- package/dist/internal/api/models/ScorerExistsResponse.d.ts +8 -0
- package/dist/internal/api/models/ScorerExistsResponse.d.ts.map +1 -0
- package/dist/internal/api/models/ScorerExistsResponse.js +6 -0
- package/dist/internal/api/models/ScorerExistsResponse.js.map +1 -0
- package/dist/internal/api/models/ScoringResult.d.ts +17 -0
- package/dist/internal/api/models/ScoringResult.d.ts.map +1 -0
- package/dist/internal/api/models/ScoringResult.js +6 -0
- package/dist/internal/api/models/ScoringResult.js.map +1 -0
- package/dist/internal/api/models/TraceEvaluationRun.d.ts +18 -0
- package/dist/internal/api/models/TraceEvaluationRun.d.ts.map +1 -0
- package/dist/internal/api/models/TraceEvaluationRun.js +6 -0
- package/dist/internal/api/models/TraceEvaluationRun.js.map +1 -0
- package/dist/internal/api/models.d.ts +24 -0
- package/dist/internal/api/models.d.ts.map +1 -0
- package/dist/internal/api/models.js +24 -0
- package/dist/internal/api/models.js.map +1 -0
- package/dist/scorers/api-scorer.d.ts +23 -0
- package/dist/scorers/api-scorer.d.ts.map +1 -0
- package/dist/scorers/api-scorer.js +36 -0
- package/dist/scorers/api-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/answer-correctness-scorer.d.ts +11 -0
- package/dist/scorers/api_scorers/answer-correctness-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/answer-correctness-scorer.js +20 -0
- package/dist/scorers/api_scorers/answer-correctness-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/answer-relevancy-scorer.d.ts +11 -0
- package/dist/scorers/api_scorers/answer-relevancy-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/answer-relevancy-scorer.js +19 -0
- package/dist/scorers/api_scorers/answer-relevancy-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/faithfulness-scorer.d.ts +11 -0
- package/dist/scorers/api_scorers/faithfulness-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/faithfulness-scorer.js +20 -0
- package/dist/scorers/api_scorers/faithfulness-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/index.d.ts +2 -0
- package/dist/scorers/api_scorers/index.d.ts.map +1 -0
- package/dist/scorers/api_scorers/index.js +2 -0
- package/dist/scorers/api_scorers/index.js.map +1 -0
- package/dist/scorers/api_scorers/instruction-adherence-scorer.d.ts +11 -0
- package/dist/scorers/api_scorers/instruction-adherence-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/instruction-adherence-scorer.js +20 -0
- package/dist/scorers/api_scorers/instruction-adherence-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts +48 -0
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js +133 -0
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/index.d.ts +5 -0
- package/dist/scorers/api_scorers/prompt_scorer/index.d.ts.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/index.js +5 -0
- package/dist/scorers/api_scorers/prompt_scorer/index.js.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.d.ts +14 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.d.ts.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js +44 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.d.ts +7 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js +9 -0
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.d.ts +6 -0
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.d.ts.map +1 -0
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js +8 -0
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js.map +1 -0
- package/dist/scorers/base-scorer.d.ts +10 -0
- package/dist/scorers/base-scorer.d.ts.map +1 -0
- package/dist/scorers/base-scorer.js +41 -0
- package/dist/scorers/base-scorer.js.map +1 -0
- package/dist/scorers/index.d.ts +4 -0
- package/dist/scorers/index.d.ts.map +1 -0
- package/dist/scorers/index.js +4 -0
- package/dist/scorers/index.js.map +1 -0
- package/dist/tracer/OpenTelemetryKeys.d.ts +12 -0
- package/dist/tracer/OpenTelemetryKeys.d.ts.map +1 -0
- package/dist/tracer/OpenTelemetryKeys.js +12 -0
- package/dist/tracer/OpenTelemetryKeys.js.map +1 -0
- package/dist/tracer/Tracer.d.ts +47 -0
- package/dist/tracer/Tracer.d.ts.map +1 -0
- package/dist/tracer/Tracer.js +235 -0
- package/dist/tracer/Tracer.js.map +1 -0
- package/dist/tracer/TracerConfiguration.d.ts +71 -0
- package/dist/tracer/TracerConfiguration.d.ts.map +1 -0
- package/dist/tracer/TracerConfiguration.js +111 -0
- package/dist/tracer/TracerConfiguration.js.map +1 -0
- package/dist/tracer/exporters/JudgmentSpanExporter.d.ts +36 -0
- package/dist/tracer/exporters/JudgmentSpanExporter.d.ts.map +1 -0
- package/dist/tracer/exporters/JudgmentSpanExporter.js +72 -0
- package/dist/tracer/exporters/JudgmentSpanExporter.js.map +1 -0
- package/dist/tracer/exporters/NoOpSpanExporter.d.ts +12 -0
- package/dist/tracer/exporters/NoOpSpanExporter.d.ts.map +1 -0
- package/dist/tracer/exporters/NoOpSpanExporter.js +16 -0
- package/dist/tracer/exporters/NoOpSpanExporter.js.map +1 -0
- package/dist/tracer/exporters/index.d.ts +3 -0
- package/dist/tracer/exporters/index.d.ts.map +1 -0
- package/dist/tracer/exporters/index.js +3 -0
- package/dist/tracer/exporters/index.js.map +1 -0
- package/dist/tracer/index.d.ts +5 -0
- package/dist/tracer/index.d.ts.map +1 -0
- package/dist/tracer/index.js +5 -0
- package/dist/tracer/index.js.map +1 -0
- package/dist/umd.d.ts +2 -0
- package/dist/umd.d.ts.map +1 -0
- package/dist/umd.js +2 -0
- package/dist/umd.js.map +1 -0
- package/dist/utils/annotate.d.ts +2 -0
- package/dist/utils/annotate.d.ts.map +1 -0
- package/dist/utils/annotate.js +27 -0
- package/dist/utils/annotate.js.map +1 -0
- package/dist/utils/index.d.ts +5 -0
- package/dist/utils/index.d.ts.map +1 -0
- package/dist/utils/index.js +5 -0
- package/dist/utils/index.js.map +1 -0
- package/dist/utils/logger.d.ts +27 -0
- package/dist/utils/logger.d.ts.map +1 -0
- package/dist/utils/logger.js +70 -0
- package/dist/utils/logger.js.map +1 -0
- package/dist/utils/types.d.ts +4 -0
- package/dist/utils/types.d.ts.map +1 -0
- package/dist/utils/types.js +2 -0
- package/dist/utils/types.js.map +1 -0
- package/dist/utils/validation.d.ts +11 -0
- package/dist/utils/validation.d.ts.map +1 -0
- package/dist/utils/validation.js +16 -0
- package/dist/utils/validation.js.map +1 -0
- package/dist/version.d.ts +2 -0
- package/dist/version.d.ts.map +1 -0
- package/dist/version.js +2 -0
- package/dist/version.js.map +1 -0
- package/package.json +57 -108
- package/LICENSE.md +0 -202
- package/README.md +0 -372
- package/dist/cjs/clients.js +0 -78
- package/dist/cjs/clients.js.map +0 -1
- package/dist/cjs/common/integrations/langgraph.js +0 -471
- package/dist/cjs/common/integrations/langgraph.js.map +0 -1
- package/dist/cjs/common/logger-instance.js +0 -64
- package/dist/cjs/common/logger-instance.js.map +0 -1
- package/dist/cjs/common/logger.js +0 -225
- package/dist/cjs/common/logger.js.map +0 -1
- package/dist/cjs/common/token-costs.js +0 -95
- package/dist/cjs/common/token-costs.js.map +0 -1
- package/dist/cjs/common/tracer.js +0 -1099
- package/dist/cjs/common/tracer.js.map +0 -1
- package/dist/cjs/constants.js +0 -347
- package/dist/cjs/constants.js.map +0 -1
- package/dist/cjs/data/datasets/eval-dataset-client.js +0 -394
- package/dist/cjs/data/datasets/eval-dataset-client.js.map +0 -1
- package/dist/cjs/data/datasets/eval-dataset.js +0 -405
- package/dist/cjs/data/datasets/eval-dataset.js.map +0 -1
- package/dist/cjs/data/example.js +0 -159
- package/dist/cjs/data/example.js.map +0 -1
- package/dist/cjs/data/result.js +0 -83
- package/dist/cjs/data/result.js.map +0 -1
- package/dist/cjs/e2etests/eval-operations.test.js +0 -282
- package/dist/cjs/e2etests/eval-operations.test.js.map +0 -1
- package/dist/cjs/e2etests/judgee-traces.test.js +0 -278
- package/dist/cjs/e2etests/judgee-traces.test.js.map +0 -1
- package/dist/cjs/evaluation-run.js +0 -136
- package/dist/cjs/evaluation-run.js.map +0 -1
- package/dist/cjs/exporters/otel-exporter.js +0 -352
- package/dist/cjs/exporters/otel-exporter.js.map +0 -1
- package/dist/cjs/index.js +0 -78
- package/dist/cjs/index.js.map +0 -1
- package/dist/cjs/judges/index.js +0 -237
- package/dist/cjs/judges/index.js.map +0 -1
- package/dist/cjs/judgment-client.js +0 -798
- package/dist/cjs/judgment-client.js.map +0 -1
- package/dist/cjs/rules.js +0 -322
- package/dist/cjs/rules.js.map +0 -1
- package/dist/cjs/run-evaluation.js +0 -700
- package/dist/cjs/run-evaluation.js.map +0 -1
- package/dist/cjs/scorers/api-scorer.js +0 -304
- package/dist/cjs/scorers/api-scorer.js.map +0 -1
- package/dist/cjs/scorers/base-scorer.js +0 -293
- package/dist/cjs/scorers/base-scorer.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-correctness/answer-correctness.js +0 -562
- package/dist/cjs/scorers/metrics/answer-correctness/answer-correctness.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-correctness/index.js +0 -19
- package/dist/cjs/scorers/metrics/answer-correctness/index.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-correctness/prompts.js +0 -175
- package/dist/cjs/scorers/metrics/answer-correctness/prompts.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-relevancy/answer-relevancy.js +0 -514
- package/dist/cjs/scorers/metrics/answer-relevancy/answer-relevancy.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-relevancy/index.js +0 -19
- package/dist/cjs/scorers/metrics/answer-relevancy/index.js.map +0 -1
- package/dist/cjs/scorers/metrics/answer-relevancy/prompts.js +0 -179
- package/dist/cjs/scorers/metrics/answer-relevancy/prompts.js.map +0 -1
- package/dist/cjs/scorers/metrics/faithfulness/faithfulness.js +0 -512
- package/dist/cjs/scorers/metrics/faithfulness/faithfulness.js.map +0 -1
- package/dist/cjs/scorers/metrics/faithfulness/index.js +0 -19
- package/dist/cjs/scorers/metrics/faithfulness/index.js.map +0 -1
- package/dist/cjs/scorers/metrics/faithfulness/prompts.js +0 -232
- package/dist/cjs/scorers/metrics/faithfulness/prompts.js.map +0 -1
- package/dist/cjs/scorers/metrics/hallucination/hallucination.js +0 -386
- package/dist/cjs/scorers/metrics/hallucination/hallucination.js.map +0 -1
- package/dist/cjs/scorers/metrics/hallucination/index.js +0 -11
- package/dist/cjs/scorers/metrics/hallucination/index.js.map +0 -1
- package/dist/cjs/scorers/metrics/hallucination/prompts.js +0 -106
- package/dist/cjs/scorers/metrics/hallucination/prompts.js.map +0 -1
- package/dist/cjs/scorers/metrics/instruction-adherence/index.js +0 -19
- package/dist/cjs/scorers/metrics/instruction-adherence/index.js.map +0 -1
- package/dist/cjs/scorers/metrics/instruction-adherence/instruction-adherence.js +0 -378
- package/dist/cjs/scorers/metrics/instruction-adherence/instruction-adherence.js.map +0 -1
- package/dist/cjs/scorers/metrics/instruction-adherence/prompts.js +0 -124
- package/dist/cjs/scorers/metrics/instruction-adherence/prompts.js.map +0 -1
- package/dist/esm/clients.js +0 -47
- package/dist/esm/clients.js.map +0 -1
- package/dist/esm/common/integrations/langgraph.js +0 -444
- package/dist/esm/common/integrations/langgraph.js.map +0 -1
- package/dist/esm/common/logger-instance.js +0 -59
- package/dist/esm/common/logger-instance.js.map +0 -1
- package/dist/esm/common/logger.js +0 -208
- package/dist/esm/common/logger.js.map +0 -1
- package/dist/esm/common/token-costs.js +0 -91
- package/dist/esm/common/token-costs.js.map +0 -1
- package/dist/esm/common/tracer.js +0 -1093
- package/dist/esm/common/tracer.js.map +0 -1
- package/dist/esm/constants.js +0 -341
- package/dist/esm/constants.js.map +0 -1
- package/dist/esm/data/datasets/eval-dataset-client.js +0 -387
- package/dist/esm/data/datasets/eval-dataset-client.js.map +0 -1
- package/dist/esm/data/datasets/eval-dataset.js +0 -375
- package/dist/esm/data/datasets/eval-dataset.js.map +0 -1
- package/dist/esm/data/example.js +0 -154
- package/dist/esm/data/example.js.map +0 -1
- package/dist/esm/data/result.js +0 -78
- package/dist/esm/data/result.js.map +0 -1
- package/dist/esm/e2etests/eval-operations.test.js +0 -254
- package/dist/esm/e2etests/eval-operations.test.js.map +0 -1
- package/dist/esm/e2etests/judgee-traces.test.js +0 -253
- package/dist/esm/e2etests/judgee-traces.test.js.map +0 -1
- package/dist/esm/evaluation-run.js +0 -132
- package/dist/esm/evaluation-run.js.map +0 -1
- package/dist/esm/exporters/otel-exporter.js +0 -348
- package/dist/esm/exporters/otel-exporter.js.map +0 -1
- package/dist/esm/index.js +0 -20
- package/dist/esm/index.js.map +0 -1
- package/dist/esm/judges/index.js +0 -205
- package/dist/esm/judges/index.js.map +0 -1
- package/dist/esm/judgment-client.js +0 -768
- package/dist/esm/judgment-client.js.map +0 -1
- package/dist/esm/rules.js +0 -314
- package/dist/esm/rules.js.map +0 -1
- package/dist/esm/run-evaluation.js +0 -681
- package/dist/esm/run-evaluation.js.map +0 -1
- package/dist/esm/scorers/api-scorer.js +0 -286
- package/dist/esm/scorers/api-scorer.js.map +0 -1
- package/dist/esm/scorers/base-scorer.js +0 -287
- package/dist/esm/scorers/base-scorer.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-correctness/answer-correctness.js +0 -558
- package/dist/esm/scorers/metrics/answer-correctness/answer-correctness.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-correctness/index.js +0 -3
- package/dist/esm/scorers/metrics/answer-correctness/index.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-correctness/prompts.js +0 -171
- package/dist/esm/scorers/metrics/answer-correctness/prompts.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-relevancy/answer-relevancy.js +0 -510
- package/dist/esm/scorers/metrics/answer-relevancy/answer-relevancy.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-relevancy/index.js +0 -3
- package/dist/esm/scorers/metrics/answer-relevancy/index.js.map +0 -1
- package/dist/esm/scorers/metrics/answer-relevancy/prompts.js +0 -175
- package/dist/esm/scorers/metrics/answer-relevancy/prompts.js.map +0 -1
- package/dist/esm/scorers/metrics/faithfulness/faithfulness.js +0 -508
- package/dist/esm/scorers/metrics/faithfulness/faithfulness.js.map +0 -1
- package/dist/esm/scorers/metrics/faithfulness/index.js +0 -3
- package/dist/esm/scorers/metrics/faithfulness/index.js.map +0 -1
- package/dist/esm/scorers/metrics/faithfulness/prompts.js +0 -228
- package/dist/esm/scorers/metrics/faithfulness/prompts.js.map +0 -1
- package/dist/esm/scorers/metrics/hallucination/hallucination.js +0 -382
- package/dist/esm/scorers/metrics/hallucination/hallucination.js.map +0 -1
- package/dist/esm/scorers/metrics/hallucination/index.js +0 -3
- package/dist/esm/scorers/metrics/hallucination/index.js.map +0 -1
- package/dist/esm/scorers/metrics/hallucination/prompts.js +0 -102
- package/dist/esm/scorers/metrics/hallucination/prompts.js.map +0 -1
- package/dist/esm/scorers/metrics/instruction-adherence/index.js +0 -3
- package/dist/esm/scorers/metrics/instruction-adherence/index.js.map +0 -1
- package/dist/esm/scorers/metrics/instruction-adherence/instruction-adherence.js +0 -374
- package/dist/esm/scorers/metrics/instruction-adherence/instruction-adherence.js.map +0 -1
- package/dist/esm/scorers/metrics/instruction-adherence/prompts.js +0 -120
- package/dist/esm/scorers/metrics/instruction-adherence/prompts.js.map +0 -1
- package/dist/types/clients.d.ts +0 -6
- package/dist/types/common/integrations/langgraph.d.ts +0 -39
- package/dist/types/common/logger-instance.d.ts +0 -3
- package/dist/types/common/logger.d.ts +0 -54
- package/dist/types/common/token-costs.d.ts +0 -36
- package/dist/types/common/tracer.d.ts +0 -249
- package/dist/types/constants.d.ts +0 -51
- package/dist/types/data/datasets/eval-dataset-client.d.ts +0 -44
- package/dist/types/data/datasets/eval-dataset.d.ts +0 -45
- package/dist/types/data/example.d.ts +0 -82
- package/dist/types/data/result.d.ts +0 -51
- package/dist/types/e2etests/eval-operations.test.d.ts +0 -5
- package/dist/types/e2etests/judgee-traces.test.d.ts +0 -5
- package/dist/types/evaluation-run.d.ts +0 -44
- package/dist/types/exporters/otel-exporter.d.ts +0 -16
- package/dist/types/index.d.ts +0 -11
- package/dist/types/judges/index.d.ts +0 -50
- package/dist/types/judgment-client.d.ts +0 -149
- package/dist/types/rules.d.ts +0 -120
- package/dist/types/run-evaluation.d.ts +0 -78
- package/dist/types/scorers/api-scorer.d.ts +0 -84
- package/dist/types/scorers/base-scorer.d.ts +0 -162
- package/dist/types/scorers/metrics/answer-correctness/answer-correctness.d.ts +0 -80
- package/dist/types/scorers/metrics/answer-correctness/index.d.ts +0 -2
- package/dist/types/scorers/metrics/answer-correctness/prompts.d.ts +0 -71
- package/dist/types/scorers/metrics/answer-relevancy/answer-relevancy.d.ts +0 -74
- package/dist/types/scorers/metrics/answer-relevancy/index.d.ts +0 -2
- package/dist/types/scorers/metrics/answer-relevancy/prompts.d.ts +0 -71
- package/dist/types/scorers/metrics/faithfulness/faithfulness.d.ts +0 -73
- package/dist/types/scorers/metrics/faithfulness/index.d.ts +0 -2
- package/dist/types/scorers/metrics/faithfulness/prompts.d.ts +0 -94
- package/dist/types/scorers/metrics/hallucination/hallucination.d.ts +0 -63
- package/dist/types/scorers/metrics/hallucination/index.d.ts +0 -3
- package/dist/types/scorers/metrics/hallucination/prompts.d.ts +0 -63
- package/dist/types/scorers/metrics/instruction-adherence/index.d.ts +0 -2
- package/dist/types/scorers/metrics/instruction-adherence/instruction-adherence.d.ts +0 -63
- package/dist/types/scorers/metrics/instruction-adherence/prompts.d.ts +0 -78
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"base-scorer.js","sourceRoot":"","sources":["../../../src/scorers/base-scorer.ts"],"names":[],"mappings":";;;;;;;;;;;;AAEA,kDAA+D;AAC/D,6DAA+D;AAC/D,mDAA2C;AA4B3C;;GAEG;AACH,MAAsB,iBAAiB;IAErC,IAAI,SAAS,KAAa,OAAO,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,6BAA6B;IAW3E,YAAY,IAAY,EAAE,SAAiB,EAAE,mBAAyC,EAAE,cAAuB,KAAK,EAAE,aAAsB,IAAI,EAAE,eAAwB,IAAI,EAAE,iBAA0B,IAAI;QAF9M,mBAAc,GAAa,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC,CAAC,0BAA0B;QAG/E,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,mBAAmB,GAAG,mBAAmB,CAAC;QAC/C,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;IACvC,CAAC;IAED;;OAEG;IACH,YAAY;QACV,IAAI,IAAI,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC7B,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC;IACtC,CAAC;IAED;;OAEG;IACH,iBAAiB;QACf,uCAAuC;QACvC,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,gCAAiB,CAAC,CAAC,IAAI,CACpD,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAC3D,CAAC;QAEF,IAAI,WAAW,EAAE,CAAC;YAChB,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBACvB,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,6CAA6C,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAC3G,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBAC7C,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,kCAAkC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAChG,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACH,MAAM;QACJ,MAAM,MAAM,GAAwB;YAClC,UAAU,EAAE,IAAI,CAAC,IAAI;YACrB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,mBAAmB,EAAE,IAAI,CAAC,mBAAmB;YAC7C,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,cAAc,EAAE,IAAI,CAAC,cAAc;SACpC,CAAC;QAEF,OAAO,MAAM,CAAC;IAChB,CAAC;IAEK,eAAe,CAAC,OAAgB;;YACpC,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAClE,CAAC;KAAA;CACF;AA3ED,8CA2EC;AAED;;GAEG;AACH,MAAsB,cAAc;IAmBlC,YACE,IAAY,EACZ,SAAiB,EACjB,mBAAyC,EACzC,iBAA0B,IAAI,EAC9B,aAAsB,IAAI,EAC1B,cAAuB,KAAK,EAC5B,eAAwB,IAAI;QAT9B,mBAAc,GAAa,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC,CAAC,0BAA0B;QAW9E,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,CAAC,6BAA6B;QACpD,IAAI,CAAC,SAAS,GAAG,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC;QAC/C,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;QACrC,IAAI,CAAC,mBAAmB,GAAG,mBAAmB,CAAC;QAC/C,IAAI,CAAC,iBAAiB,EAAE,CAAC;IAC3B,CAAC;IAED;;OAEG;IACH,YAAY;QACV,IAAI,IAAI,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC7B,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC;IAClE,CAAC;IAED;;;OAGG;IACO,aAAa;QACrB,OAAO,IAAI,CAAC,YAAY,EAAE,CAAC;IAC7B,CAAC;IAED;;OAEG;IACH,iBAAiB;QACf,uCAAuC;QACvC,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,gCAAiB,CAAC,CAAC,IAAI,CACpD,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAC3D,CAAC;QAEF,IAAI,WAAW,EAAE,CAAC;YAChB,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBACvB,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,6CAA6C,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAC3G,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBAC7C,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,kCAAkC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAChG,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACH,MAAM;QACJ,OAAO;YACL,UAAU,EAAE,IAAI,CAAC,IAAI;YACrB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,cAAc,EAAE,IAAI,CAAC,cAAc;YACnC,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,mBAAmB,EAAE,IAAI,CAAC,mBAAmB;SAC9C,CAAC;IACJ,CAAC;IAED;;;OAGG;IACO,mBAAmB,CAAC,OAAgB;QAC5C,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;YACxC,IAAI,KAAK,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;YAClE,CAAC;iBAAM,IAAI,KAAK,KAAK,cAAc,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC;gBAC7D,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACzE,CAAC;iBAAM,IAAI,KAAK,KAAK,gBAAgB,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;gBACjE,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;YAC3E,CAAC;iBAAM,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;gBACxF,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;YACvF,CAAC;iBAAM,IAAI,KAAK,KAAK,kBAAkB,IAAI,CAAC,CAAC,OAAO,CAAC,gBAAgB,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,EAAE,CAAC;gBACnH,MAAM,IAAI,KAAK,CAAC,4EAA4E,CAAC,CAAC;YAChG,CAAC;QACH,CAAC;IACH,CAAC;IAQD;;;OAGG;IACH,IAAI,IAAI;QACN,OAAO,IAAI,CAAC,IAAI,CAAC;IACnB,CAAC;IAED;;;;;;;;OAQG;IACO,oBAAoB,CAC5B,KAAa,EACb,YAAoB,EACpB,gBAAwB;QAExB,IAAI,CAAC;YACH,MAAM,UAAU,GAAG,IAAA,oCAAmB,EAAC,KAAK,EAAE,YAAY,EAAE,gBAAgB,CAAC,CAAC;YAC9E,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,IAAA,gBAAI,EAAC,kCAAkC,CAAC,EAAE,CAAC,CAAC;YAC5C,OAAO,CAAC,CAAC;QACX,CAAC;IACH,CAAC;CACF;AA5JD,wCA4JC;AAED;;GAEG;AACH,MAAa,aAAa;IAoBxB,YAAY,MAAW;QACrB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;QACxB,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,UAAU,CAAC,CAAC,6BAA6B;QACrF,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAClC,IAAI,CAAC,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QAC1B,IAAI,CAAC,eAAe,GAAG,MAAM,CAAC,eAAe,CAAC;QAC9C,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;QAC5B,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC9B,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,gBAAgB,CAAC;QAChD,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,WAAW,CAAC;QACtC,IAAI,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;QACpC,IAAI,CAAC,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACxC,IAAI,CAAC,cAAc,GAAG,MAAM,CAAC,cAAc,CAAC;QAC5C,IAAI,CAAC,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QAC1B,IAAI,CAAC,eAAe,GAAG,MAAM,CAAC,eAAe,CAAC;QAC9C,IAAI,CAAC,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACxC,IAAI,CAAC,mBAAmB,GAAG,MAAM,CAAC,mBAAmB,CAAC;QACtD,IAAI,CAAC,cAAc,GAAG,MAAM,CAAC,cAAc,CAAC;IAC9C,CAAC;IAED;;OAEG;IACH,YAAY;QACV,IAAI,IAAI,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC7B,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC;IACtC,CAAC;IAED;;OAEG;IACH,kBAAkB,CAAC,cAAuB,IAAI;QAC5C,+DAA+D;QAC/D,8CAA8C;QAC9C,IAAI,WAAW,EAAE,CAAC;YAChB,4BAA4B;YAC5B,OAAO,IAAI,CAAC,MAA2B,CAAC;QAC1C,CAAC;aAAM,CAAC;YACN,8BAA8B;YAC9B,OAAO,IAAI,CAAC,MAAwB,CAAC;QACvC,CAAC;IACH,CAAC;IAED;;OAEG;IACH,iBAAiB;QACf,uCAAuC;QACvC,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAAC,gCAAiB,CAAC,CAAC,IAAI,CACpD,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAC3D,CAAC;QAEF,IAAI,WAAW,EAAE,CAAC;YAChB,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBACvB,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,6CAA6C,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAC3G,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,EAAE,CAAC;gBAC7C,MAAM,IAAI,KAAK,CAAC,iBAAiB,IAAI,CAAC,IAAI,kCAAkC,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC;YAChG,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACH,MAAM;QACJ,OAAO;YACL,UAAU,EAAE,IAAI,CAAC,IAAI;YACrB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,UAAU;YAC3B,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,cAAc,EAAE,IAAI,CAAC,cAAc;YACnC,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,eAAe,EAAE,IAAI,CAAC,eAAe;YACrC,YAAY,EAAE,IAAI,CAAC,YAAY;YAC/B,mBAAmB,EAAE,IAAI,CAAC,mBAAmB;YAC7C,cAAc,EAAE,IAAI,CAAC,cAAc;SACpC,CAAC;IACJ,CAAC;CACF;AA7GD,sCA6GC"}
|
|
@@ -1,562 +0,0 @@
|
|
|
1
|
-
"use strict";
|
|
2
|
-
var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
|
|
3
|
-
function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
|
|
4
|
-
return new (P || (P = Promise))(function (resolve, reject) {
|
|
5
|
-
function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
|
|
6
|
-
function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
|
|
7
|
-
function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
|
|
8
|
-
step((generator = generator.apply(thisArg, _arguments || [])).next());
|
|
9
|
-
});
|
|
10
|
-
};
|
|
11
|
-
Object.defineProperty(exports, "__esModule", { value: true });
|
|
12
|
-
exports.AnswerCorrectnessScorer = void 0;
|
|
13
|
-
const logger_js_1 = require("../../../common/logger.js");
|
|
14
|
-
const base_scorer_js_1 = require("../../base-scorer.js");
|
|
15
|
-
const constants_js_1 = require("../../../constants.js");
|
|
16
|
-
const prompts_js_1 = require("./prompts.js");
|
|
17
|
-
const index_js_1 = require("../../../judges/index.js");
|
|
18
|
-
// Required parameters for this scorer
|
|
19
|
-
const requiredParams = ['input', 'actualOutput', 'expectedOutput'];
|
|
20
|
-
/**
|
|
21
|
-
* Answer Correctness Scorer
|
|
22
|
-
*
|
|
23
|
-
* This scorer evaluates whether the actual output correctly represents
|
|
24
|
-
* the expected output by extracting statements from the expected output
|
|
25
|
-
* and checking if they are correctly represented in the actual output.
|
|
26
|
-
*/
|
|
27
|
-
class AnswerCorrectnessScorer extends base_scorer_js_1.JudgevalScorer {
|
|
28
|
-
/**
|
|
29
|
-
* Constructor for AnswerCorrectnessScorer
|
|
30
|
-
* @param threshold Minimum score to consider the evaluation successful (default: 0.5)
|
|
31
|
-
* @param model LLM to use for evaluation (string or Judge instance)
|
|
32
|
-
* @param include_reason Whether to generate a reason for the score
|
|
33
|
-
* @param async_mode Whether to use asynchronous evaluation
|
|
34
|
-
* @param strict_mode If true, sets threshold to 1.0 (requiring perfect match)
|
|
35
|
-
* @param verbose_mode Enables detailed logging
|
|
36
|
-
* @param user Optional user identifier for the LLM
|
|
37
|
-
* @param additional_metadata Additional metadata to include in the result
|
|
38
|
-
*/
|
|
39
|
-
constructor(threshold = 0.5, model, include_reason = true, async_mode = true, strict_mode = false, verbose_mode = true, user, additional_metadata) {
|
|
40
|
-
super(constants_js_1.APIScorer.ANSWER_CORRECTNESS, strict_mode ? 1.0 : threshold, additional_metadata, include_reason, async_mode, strict_mode, verbose_mode);
|
|
41
|
-
(0, logger_js_1.info)(`Initializing AnswerCorrectnessScorer with threshold=${this.threshold}, model=${model}, strict_mode=${strict_mode}`);
|
|
42
|
-
const { judge, usingNativeModel } = (0, index_js_1.createJudge)(model, user);
|
|
43
|
-
this.model = judge;
|
|
44
|
-
this.usingNativeModel = usingNativeModel;
|
|
45
|
-
this.evaluation_model = this.model.getModelName();
|
|
46
|
-
(0, logger_js_1.log)(`Using model: ${this.evaluation_model}`);
|
|
47
|
-
// Set required fields for this scorer
|
|
48
|
-
this.requiredFields = ['input', 'actualOutput', 'expectedOutput'];
|
|
49
|
-
}
|
|
50
|
-
/**
|
|
51
|
-
* Get statements from expected output asynchronously
|
|
52
|
-
*/
|
|
53
|
-
_aGetStatements(expectedOutput) {
|
|
54
|
-
return __awaiter(this, void 0, void 0, function* () {
|
|
55
|
-
(0, logger_js_1.log)("Getting statements asynchronously");
|
|
56
|
-
// Handle string array
|
|
57
|
-
const expectedOutputStr = Array.isArray(expectedOutput) ? expectedOutput.join('\n') : expectedOutput;
|
|
58
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.deduceStatements(expectedOutputStr);
|
|
59
|
-
try {
|
|
60
|
-
const response = yield this.model.aGenerate(prompt);
|
|
61
|
-
// Parse the response
|
|
62
|
-
try {
|
|
63
|
-
const jsonResponse = JSON.parse(response);
|
|
64
|
-
const parsed = prompts_js_1.StatementsSchema.safeParse(jsonResponse);
|
|
65
|
-
if (parsed.success) {
|
|
66
|
-
return parsed.data.statements;
|
|
67
|
-
}
|
|
68
|
-
else {
|
|
69
|
-
// Fallback to direct access if schema validation fails
|
|
70
|
-
(0, logger_js_1.warn)("Schema validation failed, falling back to raw response parsing");
|
|
71
|
-
if (jsonResponse.statements && Array.isArray(jsonResponse.statements)) {
|
|
72
|
-
return jsonResponse.statements;
|
|
73
|
-
}
|
|
74
|
-
}
|
|
75
|
-
}
|
|
76
|
-
catch (parseError) {
|
|
77
|
-
(0, logger_js_1.warn)(`Error parsing JSON response: ${parseError}`);
|
|
78
|
-
// Try to extract JSON from the response text
|
|
79
|
-
const jsonMatch = response.match(/\{[\s\S]*\}/);
|
|
80
|
-
if (jsonMatch) {
|
|
81
|
-
try {
|
|
82
|
-
const extractedJson = JSON.parse(jsonMatch[0]);
|
|
83
|
-
if (extractedJson.statements && Array.isArray(extractedJson.statements)) {
|
|
84
|
-
return extractedJson.statements;
|
|
85
|
-
}
|
|
86
|
-
}
|
|
87
|
-
catch (e) {
|
|
88
|
-
(0, logger_js_1.error)(`Failed to extract JSON from response: ${e}`);
|
|
89
|
-
}
|
|
90
|
-
}
|
|
91
|
-
}
|
|
92
|
-
// If all parsing attempts fail, return empty array
|
|
93
|
-
(0, logger_js_1.error)("Failed to parse statements from model response");
|
|
94
|
-
return [];
|
|
95
|
-
}
|
|
96
|
-
catch (e) {
|
|
97
|
-
(0, logger_js_1.error)(`Error getting statements: ${e}`);
|
|
98
|
-
return [];
|
|
99
|
-
}
|
|
100
|
-
});
|
|
101
|
-
}
|
|
102
|
-
/**
|
|
103
|
-
* Get statements from expected output synchronously
|
|
104
|
-
*/
|
|
105
|
-
_getStatements(expectedOutput) {
|
|
106
|
-
// Handle string array
|
|
107
|
-
const expectedOutputStr = Array.isArray(expectedOutput) ? expectedOutput.join('\n') : expectedOutput;
|
|
108
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.deduceStatements(expectedOutputStr);
|
|
109
|
-
try {
|
|
110
|
-
const response = this.model.generate(prompt);
|
|
111
|
-
// Parse the response
|
|
112
|
-
try {
|
|
113
|
-
const jsonResponse = JSON.parse(response);
|
|
114
|
-
const parsed = prompts_js_1.StatementsSchema.safeParse(jsonResponse);
|
|
115
|
-
if (parsed.success) {
|
|
116
|
-
return parsed.data.statements;
|
|
117
|
-
}
|
|
118
|
-
else {
|
|
119
|
-
// Fallback to direct access if schema validation fails
|
|
120
|
-
(0, logger_js_1.warn)("Schema validation failed, falling back to raw response parsing");
|
|
121
|
-
if (jsonResponse.statements && Array.isArray(jsonResponse.statements)) {
|
|
122
|
-
return jsonResponse.statements;
|
|
123
|
-
}
|
|
124
|
-
}
|
|
125
|
-
}
|
|
126
|
-
catch (parseError) {
|
|
127
|
-
(0, logger_js_1.warn)(`Error parsing JSON response: ${parseError}`);
|
|
128
|
-
// Try to extract JSON from the response text
|
|
129
|
-
const jsonMatch = response.match(/\{[\s\S]*\}/);
|
|
130
|
-
if (jsonMatch) {
|
|
131
|
-
try {
|
|
132
|
-
const extractedJson = JSON.parse(jsonMatch[0]);
|
|
133
|
-
if (extractedJson.statements && Array.isArray(extractedJson.statements)) {
|
|
134
|
-
return extractedJson.statements;
|
|
135
|
-
}
|
|
136
|
-
}
|
|
137
|
-
catch (e) {
|
|
138
|
-
(0, logger_js_1.error)(`Failed to extract JSON from response: ${e}`);
|
|
139
|
-
}
|
|
140
|
-
}
|
|
141
|
-
}
|
|
142
|
-
// If all parsing attempts fail, return empty array
|
|
143
|
-
(0, logger_js_1.error)("Failed to parse statements from model response");
|
|
144
|
-
return [];
|
|
145
|
-
}
|
|
146
|
-
catch (e) {
|
|
147
|
-
(0, logger_js_1.error)(`Error getting statements: ${e}`);
|
|
148
|
-
return [];
|
|
149
|
-
}
|
|
150
|
-
}
|
|
151
|
-
/**
|
|
152
|
-
* Get verdicts for statements against actual output asynchronously
|
|
153
|
-
*/
|
|
154
|
-
_aGetVerdicts(actualOutput) {
|
|
155
|
-
return __awaiter(this, void 0, void 0, function* () {
|
|
156
|
-
(0, logger_js_1.log)("Getting verdicts asynchronously");
|
|
157
|
-
if (!this.statements || this.statements.length === 0) {
|
|
158
|
-
(0, logger_js_1.warn)("No statements to evaluate");
|
|
159
|
-
return [];
|
|
160
|
-
}
|
|
161
|
-
// Handle string array
|
|
162
|
-
const actualOutputStr = Array.isArray(actualOutput) ? actualOutput.join('\n') : actualOutput;
|
|
163
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.generateVerdicts(this.statements, actualOutputStr);
|
|
164
|
-
try {
|
|
165
|
-
const response = yield this.model.aGenerate(prompt);
|
|
166
|
-
// Parse the response
|
|
167
|
-
try {
|
|
168
|
-
const jsonResponse = JSON.parse(response);
|
|
169
|
-
const parsed = prompts_js_1.VerdictsSchema.safeParse(jsonResponse);
|
|
170
|
-
if (parsed.success) {
|
|
171
|
-
return parsed.data.verdicts;
|
|
172
|
-
}
|
|
173
|
-
else {
|
|
174
|
-
// Fallback to direct access if schema validation fails
|
|
175
|
-
(0, logger_js_1.warn)("Schema validation failed, falling back to raw response parsing");
|
|
176
|
-
if (jsonResponse.verdicts && Array.isArray(jsonResponse.verdicts)) {
|
|
177
|
-
return jsonResponse.verdicts.map((v) => ({
|
|
178
|
-
verdict: v.verdict,
|
|
179
|
-
reason: v.reason
|
|
180
|
-
}));
|
|
181
|
-
}
|
|
182
|
-
}
|
|
183
|
-
}
|
|
184
|
-
catch (parseError) {
|
|
185
|
-
(0, logger_js_1.warn)(`Error parsing JSON response: ${parseError}`);
|
|
186
|
-
// Try to extract JSON from the response text
|
|
187
|
-
const jsonMatch = response.match(/\{[\s\S]*\}/);
|
|
188
|
-
if (jsonMatch) {
|
|
189
|
-
try {
|
|
190
|
-
const extractedJson = JSON.parse(jsonMatch[0]);
|
|
191
|
-
if (extractedJson.verdicts && Array.isArray(extractedJson.verdicts)) {
|
|
192
|
-
return extractedJson.verdicts.map((v) => ({
|
|
193
|
-
verdict: v.verdict,
|
|
194
|
-
reason: v.reason
|
|
195
|
-
}));
|
|
196
|
-
}
|
|
197
|
-
}
|
|
198
|
-
catch (e) {
|
|
199
|
-
(0, logger_js_1.error)(`Failed to extract JSON from response: ${e}`);
|
|
200
|
-
}
|
|
201
|
-
}
|
|
202
|
-
}
|
|
203
|
-
// If all parsing attempts fail, return empty array
|
|
204
|
-
(0, logger_js_1.error)("Failed to parse verdicts from model response");
|
|
205
|
-
return [];
|
|
206
|
-
}
|
|
207
|
-
catch (e) {
|
|
208
|
-
(0, logger_js_1.error)(`Error getting verdicts: ${e}`);
|
|
209
|
-
return [];
|
|
210
|
-
}
|
|
211
|
-
});
|
|
212
|
-
}
|
|
213
|
-
/**
|
|
214
|
-
* Get verdicts for statements against actual output synchronously
|
|
215
|
-
*/
|
|
216
|
-
_getVerdicts(actualOutput) {
|
|
217
|
-
if (!this.statements || this.statements.length === 0) {
|
|
218
|
-
(0, logger_js_1.warn)("No statements to evaluate");
|
|
219
|
-
return [];
|
|
220
|
-
}
|
|
221
|
-
// Handle string array
|
|
222
|
-
const actualOutputStr = Array.isArray(actualOutput) ? actualOutput.join('\n') : actualOutput;
|
|
223
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.generateVerdicts(this.statements, actualOutputStr);
|
|
224
|
-
try {
|
|
225
|
-
const response = this.model.generate(prompt);
|
|
226
|
-
// Parse the response
|
|
227
|
-
try {
|
|
228
|
-
const jsonResponse = JSON.parse(response);
|
|
229
|
-
const parsed = prompts_js_1.VerdictsSchema.safeParse(jsonResponse);
|
|
230
|
-
if (parsed.success) {
|
|
231
|
-
return parsed.data.verdicts;
|
|
232
|
-
}
|
|
233
|
-
else {
|
|
234
|
-
// Fallback to direct access if schema validation fails
|
|
235
|
-
(0, logger_js_1.warn)("Schema validation failed, falling back to raw response parsing");
|
|
236
|
-
if (jsonResponse.verdicts && Array.isArray(jsonResponse.verdicts)) {
|
|
237
|
-
return jsonResponse.verdicts.map((v) => ({
|
|
238
|
-
verdict: v.verdict,
|
|
239
|
-
reason: v.reason
|
|
240
|
-
}));
|
|
241
|
-
}
|
|
242
|
-
}
|
|
243
|
-
}
|
|
244
|
-
catch (parseError) {
|
|
245
|
-
(0, logger_js_1.warn)(`Error parsing JSON response: ${parseError}`);
|
|
246
|
-
// Try to extract JSON from the response text
|
|
247
|
-
const jsonMatch = response.match(/\{[\s\S]*\}/);
|
|
248
|
-
if (jsonMatch) {
|
|
249
|
-
try {
|
|
250
|
-
const extractedJson = JSON.parse(jsonMatch[0]);
|
|
251
|
-
if (extractedJson.verdicts && Array.isArray(extractedJson.verdicts)) {
|
|
252
|
-
return extractedJson.verdicts.map((v) => ({
|
|
253
|
-
verdict: v.verdict,
|
|
254
|
-
reason: v.reason
|
|
255
|
-
}));
|
|
256
|
-
}
|
|
257
|
-
}
|
|
258
|
-
catch (e) {
|
|
259
|
-
(0, logger_js_1.error)(`Failed to extract JSON from response: ${e}`);
|
|
260
|
-
}
|
|
261
|
-
}
|
|
262
|
-
}
|
|
263
|
-
// If all parsing attempts fail, return empty array
|
|
264
|
-
(0, logger_js_1.error)("Failed to parse verdicts from model response");
|
|
265
|
-
return [];
|
|
266
|
-
}
|
|
267
|
-
catch (e) {
|
|
268
|
-
(0, logger_js_1.error)(`Error getting verdicts: ${e}`);
|
|
269
|
-
return [];
|
|
270
|
-
}
|
|
271
|
-
}
|
|
272
|
-
/**
|
|
273
|
-
* Get reason for the score asynchronously
|
|
274
|
-
*/
|
|
275
|
-
_aGetReason() {
|
|
276
|
-
return __awaiter(this, void 0, void 0, function* () {
|
|
277
|
-
var _a;
|
|
278
|
-
if (!this.include_reason) {
|
|
279
|
-
return undefined;
|
|
280
|
-
}
|
|
281
|
-
if (!this.verdicts || this.verdicts.length === 0) {
|
|
282
|
-
return undefined;
|
|
283
|
-
}
|
|
284
|
-
try {
|
|
285
|
-
// Get incorrect statements with their verdicts
|
|
286
|
-
const incorrectStatements = [];
|
|
287
|
-
for (let i = 0; i < this.statements.length; i++) {
|
|
288
|
-
if (i < this.verdicts.length && this.verdicts[i].verdict.toLowerCase() === "no") {
|
|
289
|
-
incorrectStatements.push([this.statements[i], this.verdicts[i].reason]);
|
|
290
|
-
}
|
|
291
|
-
}
|
|
292
|
-
if (incorrectStatements.length === 0) {
|
|
293
|
-
return "All statements in the expected output are correctly represented in the actual output.";
|
|
294
|
-
}
|
|
295
|
-
// Generate reason
|
|
296
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.generateReason(incorrectStatements, ((_a = this.score) === null || _a === void 0 ? void 0 : _a.toString()) || "0");
|
|
297
|
-
const reasonText = yield this.model.aGenerate(prompt);
|
|
298
|
-
const parsedReason = prompts_js_1.ReasonSchema.safeParse(JSON.parse(reasonText));
|
|
299
|
-
if (!parsedReason.success) {
|
|
300
|
-
(0, logger_js_1.error)(`Failed to parse reason: ${parsedReason.error}`);
|
|
301
|
-
return undefined;
|
|
302
|
-
}
|
|
303
|
-
return parsedReason.data.reason;
|
|
304
|
-
}
|
|
305
|
-
catch (err) {
|
|
306
|
-
(0, logger_js_1.error)(`Error getting reason: ${err}`);
|
|
307
|
-
return undefined;
|
|
308
|
-
}
|
|
309
|
-
});
|
|
310
|
-
}
|
|
311
|
-
/**
|
|
312
|
-
* Get reason for the score synchronously
|
|
313
|
-
*/
|
|
314
|
-
_getReason() {
|
|
315
|
-
var _a;
|
|
316
|
-
if (!this.include_reason) {
|
|
317
|
-
return undefined;
|
|
318
|
-
}
|
|
319
|
-
if (!this.verdicts || this.verdicts.length === 0) {
|
|
320
|
-
return undefined;
|
|
321
|
-
}
|
|
322
|
-
try {
|
|
323
|
-
// Get incorrect statements with their verdicts
|
|
324
|
-
const incorrectStatements = [];
|
|
325
|
-
for (let i = 0; i < this.statements.length; i++) {
|
|
326
|
-
if (i < this.verdicts.length && this.verdicts[i].verdict.toLowerCase() === "no") {
|
|
327
|
-
incorrectStatements.push([this.statements[i], this.verdicts[i].reason]);
|
|
328
|
-
}
|
|
329
|
-
}
|
|
330
|
-
if (incorrectStatements.length === 0) {
|
|
331
|
-
return "All statements in the expected output are correctly represented in the actual output.";
|
|
332
|
-
}
|
|
333
|
-
// Generate reason
|
|
334
|
-
const prompt = prompts_js_1.AnswerCorrectnessTemplate.generateReason(incorrectStatements, ((_a = this.score) === null || _a === void 0 ? void 0 : _a.toString()) || "0");
|
|
335
|
-
const reasonText = this.model.generate(prompt);
|
|
336
|
-
const parsedReason = prompts_js_1.ReasonSchema.safeParse(JSON.parse(reasonText));
|
|
337
|
-
if (!parsedReason.success) {
|
|
338
|
-
(0, logger_js_1.error)(`Failed to parse reason: ${parsedReason.error}`);
|
|
339
|
-
return undefined;
|
|
340
|
-
}
|
|
341
|
-
return parsedReason.data.reason;
|
|
342
|
-
}
|
|
343
|
-
catch (err) {
|
|
344
|
-
(0, logger_js_1.error)(`Error getting reason: ${err}`);
|
|
345
|
-
return undefined;
|
|
346
|
-
}
|
|
347
|
-
}
|
|
348
|
-
/**
|
|
349
|
-
* Compute score based on verdicts
|
|
350
|
-
*/
|
|
351
|
-
_computeScore() {
|
|
352
|
-
(0, logger_js_1.log)("Computing score");
|
|
353
|
-
// If we have no statements or verdicts due to API errors, return 0 instead of 1
|
|
354
|
-
// This ensures that when API calls fail, we don't incorrectly return a perfect score
|
|
355
|
-
if (!this.statements || this.statements.length === 0) {
|
|
356
|
-
return 0;
|
|
357
|
-
}
|
|
358
|
-
if (!this.verdicts || this.verdicts.length === 0) {
|
|
359
|
-
return 0;
|
|
360
|
-
}
|
|
361
|
-
let correctCount = 0;
|
|
362
|
-
for (const verdict of this.verdicts) {
|
|
363
|
-
if (verdict.verdict.trim().toLowerCase() === "yes") {
|
|
364
|
-
correctCount++;
|
|
365
|
-
}
|
|
366
|
-
}
|
|
367
|
-
const score = correctCount / this.verdicts.length;
|
|
368
|
-
// Match Python implementation's handling of strict_mode
|
|
369
|
-
return this.strict_mode && score < this.threshold ? 0 : score;
|
|
370
|
-
}
|
|
371
|
-
/**
|
|
372
|
-
* Check if example has required parameters
|
|
373
|
-
*/
|
|
374
|
-
_checkExampleParams(example) {
|
|
375
|
-
for (const param of this.requiredFields) {
|
|
376
|
-
if (param === 'input' && !example.input) {
|
|
377
|
-
throw new Error(`Example is missing required parameter: input`);
|
|
378
|
-
}
|
|
379
|
-
else if (param === 'actualOutput' && !example.actualOutput) {
|
|
380
|
-
throw new Error(`Example is missing required parameter: actualOutput`);
|
|
381
|
-
}
|
|
382
|
-
else if (param === 'expectedOutput' && !example.expectedOutput) {
|
|
383
|
-
throw new Error(`Example is missing required parameter: expectedOutput`);
|
|
384
|
-
}
|
|
385
|
-
else if (param === 'context' && (!example.context || !Array.isArray(example.context))) {
|
|
386
|
-
throw new Error(`Example is missing required parameter: context (must be an array)`);
|
|
387
|
-
}
|
|
388
|
-
else if (param === 'retrievalContext' && (!example.retrievalContext || !Array.isArray(example.retrievalContext))) {
|
|
389
|
-
throw new Error(`Example is missing required parameter: retrievalContext (must be an array)`);
|
|
390
|
-
}
|
|
391
|
-
}
|
|
392
|
-
}
|
|
393
|
-
/**
|
|
394
|
-
* Create verbose logs for debugging
|
|
395
|
-
*/
|
|
396
|
-
_createVerboseLogs() {
|
|
397
|
-
if (!this.verbose_mode) {
|
|
398
|
-
return '';
|
|
399
|
-
}
|
|
400
|
-
const steps = [
|
|
401
|
-
`Statements:\n${JSON.stringify(this.statements, null, 2)}`,
|
|
402
|
-
`Verdicts:\n${JSON.stringify(this.verdicts, null, 2)}`,
|
|
403
|
-
`Score: ${this.score}\nReason: ${this.reason}`
|
|
404
|
-
];
|
|
405
|
-
return steps.join('\n\n');
|
|
406
|
-
}
|
|
407
|
-
/**
|
|
408
|
-
* Score an example synchronously - this is for compatibility with the Python SDK
|
|
409
|
-
*/
|
|
410
|
-
syncScoreExample(example) {
|
|
411
|
-
(0, logger_js_1.info)("Starting example scoring (sync mode)");
|
|
412
|
-
try {
|
|
413
|
-
// Check required parameters
|
|
414
|
-
this._checkExampleParams(example);
|
|
415
|
-
// Process example
|
|
416
|
-
if (this.async_mode) {
|
|
417
|
-
throw new Error("Cannot use synchronous scoreExample with async_mode=true. Use async scoreExample instead.");
|
|
418
|
-
}
|
|
419
|
-
// Track token usage
|
|
420
|
-
let promptTokens = 0;
|
|
421
|
-
let completionTokens = 0;
|
|
422
|
-
// Get statements and track tokens
|
|
423
|
-
this.statements = this._getStatements(example.expectedOutput);
|
|
424
|
-
promptTokens += 500; // Approximate tokens for statements prompt
|
|
425
|
-
completionTokens += 100; // Approximate tokens for statements response
|
|
426
|
-
// Get verdicts and track tokens
|
|
427
|
-
this.verdicts = this._getVerdicts(example.actualOutput);
|
|
428
|
-
promptTokens += 800; // Approximate tokens for verdicts prompt
|
|
429
|
-
completionTokens += 200; // Approximate tokens for verdicts response
|
|
430
|
-
// Compute score
|
|
431
|
-
this.score = this._computeScore();
|
|
432
|
-
// Get reason if needed
|
|
433
|
-
if (this.include_reason) {
|
|
434
|
-
this.reason = this._getReason();
|
|
435
|
-
promptTokens += 300; // Approximate tokens for reason prompt
|
|
436
|
-
completionTokens += 100; // Approximate tokens for reason response
|
|
437
|
-
}
|
|
438
|
-
// Calculate evaluation cost
|
|
439
|
-
this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
|
|
440
|
-
// Set success flag
|
|
441
|
-
this.success = this.score >= this.threshold;
|
|
442
|
-
// Create verbose logs if needed
|
|
443
|
-
if (this.verbose_mode) {
|
|
444
|
-
this.verbose_logs = this._createVerboseLogs();
|
|
445
|
-
}
|
|
446
|
-
(0, logger_js_1.info)(`Scoring completed with score: ${this.score}`);
|
|
447
|
-
// Ensure all fields match the ScorerData interface
|
|
448
|
-
return {
|
|
449
|
-
name: this.type,
|
|
450
|
-
threshold: this.threshold,
|
|
451
|
-
success: this.success || false,
|
|
452
|
-
score: this.score || 0,
|
|
453
|
-
reason: this.reason !== undefined ? this.reason : null,
|
|
454
|
-
strict_mode: this.strict_mode || false,
|
|
455
|
-
evaluation_model: this.evaluation_model || null,
|
|
456
|
-
error: null,
|
|
457
|
-
evaluation_cost: this.evaluation_cost,
|
|
458
|
-
verbose_logs: this.verbose_logs ? this.verbose_logs : null,
|
|
459
|
-
additional_metadata: this.additional_metadata || {}
|
|
460
|
-
};
|
|
461
|
-
}
|
|
462
|
-
catch (error) {
|
|
463
|
-
// Handle errors
|
|
464
|
-
const errorMessage = error instanceof Error ? error.message : String(error);
|
|
465
|
-
this.error = errorMessage;
|
|
466
|
-
this.success = false;
|
|
467
|
-
return {
|
|
468
|
-
name: this.type,
|
|
469
|
-
threshold: this.threshold,
|
|
470
|
-
success: false,
|
|
471
|
-
score: 0,
|
|
472
|
-
reason: `Error during scoring: ${errorMessage}`,
|
|
473
|
-
strict_mode: this.strict_mode || false,
|
|
474
|
-
evaluation_model: this.evaluation_model || null,
|
|
475
|
-
error: errorMessage,
|
|
476
|
-
evaluation_cost: null,
|
|
477
|
-
verbose_logs: null,
|
|
478
|
-
additional_metadata: this.additional_metadata || {}
|
|
479
|
-
};
|
|
480
|
-
}
|
|
481
|
-
}
|
|
482
|
-
/**
|
|
483
|
-
* Score an example - this is the main method that should be called
|
|
484
|
-
* It will use async or sync methods based on the async_mode setting
|
|
485
|
-
*/
|
|
486
|
-
scoreExample(example) {
|
|
487
|
-
return __awaiter(this, void 0, void 0, function* () {
|
|
488
|
-
if (!this.async_mode) {
|
|
489
|
-
return this.syncScoreExample(example);
|
|
490
|
-
}
|
|
491
|
-
(0, logger_js_1.info)("Starting example scoring (async mode)");
|
|
492
|
-
try {
|
|
493
|
-
// Check required parameters
|
|
494
|
-
this._checkExampleParams(example);
|
|
495
|
-
// Track token usage
|
|
496
|
-
let promptTokens = 0;
|
|
497
|
-
let completionTokens = 0;
|
|
498
|
-
// Process example
|
|
499
|
-
this.statements = yield this._aGetStatements(example.expectedOutput);
|
|
500
|
-
promptTokens += 500; // Approximate tokens for statements prompt
|
|
501
|
-
completionTokens += 100; // Approximate tokens for statements response
|
|
502
|
-
this.verdicts = yield this._aGetVerdicts(example.actualOutput);
|
|
503
|
-
promptTokens += 800; // Approximate tokens for verdicts prompt
|
|
504
|
-
completionTokens += 200; // Approximate tokens for verdicts response
|
|
505
|
-
this.score = this._computeScore();
|
|
506
|
-
if (this.include_reason) {
|
|
507
|
-
this.reason = yield this._aGetReason();
|
|
508
|
-
promptTokens += 300; // Approximate tokens for reason prompt
|
|
509
|
-
completionTokens += 100; // Approximate tokens for reason response
|
|
510
|
-
}
|
|
511
|
-
this.success = this._successCheck();
|
|
512
|
-
if (this.verbose_mode) {
|
|
513
|
-
this.verbose_logs = this._createVerboseLogs();
|
|
514
|
-
}
|
|
515
|
-
// Calculate evaluation cost
|
|
516
|
-
this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
|
|
517
|
-
(0, logger_js_1.info)(`Scoring completed with score: ${this.score}`);
|
|
518
|
-
// Ensure all fields match the ScorerData interface
|
|
519
|
-
return {
|
|
520
|
-
name: this.type,
|
|
521
|
-
threshold: this.threshold,
|
|
522
|
-
success: this.success || false,
|
|
523
|
-
score: this.score || 0,
|
|
524
|
-
reason: this.reason !== undefined ? this.reason : null,
|
|
525
|
-
strict_mode: this.strict_mode || false,
|
|
526
|
-
evaluation_model: this.evaluation_model || null,
|
|
527
|
-
error: null,
|
|
528
|
-
evaluation_cost: this.evaluation_cost,
|
|
529
|
-
verbose_logs: this.verbose_logs ? this.verbose_logs : null,
|
|
530
|
-
additional_metadata: this.additional_metadata || {}
|
|
531
|
-
};
|
|
532
|
-
}
|
|
533
|
-
catch (error) {
|
|
534
|
-
// Handle errors
|
|
535
|
-
const errorMessage = error instanceof Error ? error.message : String(error);
|
|
536
|
-
this.error = errorMessage;
|
|
537
|
-
this.success = false;
|
|
538
|
-
return {
|
|
539
|
-
name: this.type,
|
|
540
|
-
threshold: this.threshold,
|
|
541
|
-
success: false,
|
|
542
|
-
score: 0,
|
|
543
|
-
reason: `Error during scoring: ${errorMessage}`,
|
|
544
|
-
strict_mode: this.strict_mode || false,
|
|
545
|
-
evaluation_model: this.evaluation_model || null,
|
|
546
|
-
error: errorMessage,
|
|
547
|
-
evaluation_cost: null,
|
|
548
|
-
verbose_logs: null,
|
|
549
|
-
additional_metadata: this.additional_metadata || {}
|
|
550
|
-
};
|
|
551
|
-
}
|
|
552
|
-
});
|
|
553
|
-
}
|
|
554
|
-
/**
|
|
555
|
-
* Get the name of the scorer
|
|
556
|
-
*/
|
|
557
|
-
get name() {
|
|
558
|
-
return "Answer Correctness";
|
|
559
|
-
}
|
|
560
|
-
}
|
|
561
|
-
exports.AnswerCorrectnessScorer = AnswerCorrectnessScorer;
|
|
562
|
-
//# sourceMappingURL=answer-correctness.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"answer-correctness.js","sourceRoot":"","sources":["../../../../../src/scorers/metrics/answer-correctness/answer-correctness.ts"],"names":[],"mappings":";;;;;;;;;;;;AAAA,yDAAmE;AACnE,yDAAsD;AAGtD,wDAAkD;AAClD,6CAAoH;AACpH,uDAA8D;AAG9D,sCAAsC;AACtC,MAAM,cAAc,GAAG,CAAC,OAAO,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;AAEnE;;;;;;GAMG;AACH,MAAa,uBAAwB,SAAQ,+BAAc;IAQzD;;;;;;;;;;OAUG;IACH,YACE,YAAoB,GAAG,EACvB,KAAsB,EACtB,iBAA0B,IAAI,EAC9B,aAAsB,IAAI,EAC1B,cAAuB,KAAK,EAC5B,eAAwB,IAAI,EAC5B,IAAa,EACb,mBAAyC;QAEzC,KAAK,CACH,wBAAS,CAAC,kBAAkB,EAC5B,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAC7B,mBAAmB,EACnB,cAAc,EACd,UAAU,EACV,WAAW,EACX,YAAY,CACb,CAAC;QAEF,IAAA,gBAAI,EAAC,uDAAuD,IAAI,CAAC,SAAS,WAAW,KAAK,iBAAiB,WAAW,EAAE,CAAC,CAAC;QAE1H,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,GAAG,IAAA,sBAAW,EAAC,KAAK,EAAE,IAAI,CAAC,CAAC;QAC7D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,gBAAgB,GAAG,gBAAgB,CAAC;QACzC,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC;QAElD,IAAA,eAAG,EAAC,gBAAgB,IAAI,CAAC,gBAAgB,EAAE,CAAC,CAAC;QAE7C,sCAAsC;QACtC,IAAI,CAAC,cAAc,GAAG,CAAC,OAAO,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;IACpE,CAAC;IAED;;OAEG;IACW,eAAe,CAAC,cAAiC;;YAC7D,IAAA,eAAG,EAAC,mCAAmC,CAAC,CAAC;YAEzC,sBAAsB;YACtB,MAAM,iBAAiB,GAAG,KAAK,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC;YAErG,MAAM,MAAM,GAAG,sCAAyB,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;YAE7E,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAEpD,qBAAqB;gBACrB,IAAI,CAAC;oBACH,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;oBAC1C,MAAM,MAAM,GAAG,6BAAgB,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;oBAExD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;wBACnB,OAAO,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC;oBAChC,CAAC;yBAAM,CAAC;wBACN,uDAAuD;wBACvD,IAAA,gBAAI,EAAC,gEAAgE,CAAC,CAAC;wBACvE,IAAI,YAAY,CAAC,UAAU,IAAI,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,UAAU,CAAC,EAAE,CAAC;4BACtE,OAAO,YAAY,CAAC,UAAU,CAAC;wBACjC,CAAC;oBACH,CAAC;gBACH,CAAC;gBAAC,OAAO,UAAU,EAAE,CAAC;oBACpB,IAAA,gBAAI,EAAC,gCAAgC,UAAU,EAAE,CAAC,CAAC;oBACnD,6CAA6C;oBAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;oBAChD,IAAI,SAAS,EAAE,CAAC;wBACd,IAAI,CAAC;4BACH,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;4BAC/C,IAAI,aAAa,CAAC,UAAU,IAAI,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,UAAU,CAAC,EAAE,CAAC;gCACxE,OAAO,aAAa,CAAC,UAAU,CAAC;4BAClC,CAAC;wBACH,CAAC;wBAAC,OAAO,CAAC,EAAE,CAAC;4BACX,IAAA,iBAAK,EAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;wBACtD,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,mDAAmD;gBACnD,IAAA,iBAAK,EAAC,gDAAgD,CAAC,CAAC;gBACxD,OAAO,EAAE,CAAC;YACZ,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,IAAA,iBAAK,EAAC,6BAA6B,CAAC,EAAE,CAAC,CAAC;gBACxC,OAAO,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACK,cAAc,CAAC,cAAiC;QACtD,sBAAsB;QACtB,MAAM,iBAAiB,GAAG,KAAK,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC;QAErG,MAAM,MAAM,GAAG,sCAAyB,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;QAE7E,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YAE7C,qBAAqB;YACrB,IAAI,CAAC;gBACH,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;gBAC1C,MAAM,MAAM,GAAG,6BAAgB,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;gBAExD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;oBACnB,OAAO,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC;gBAChC,CAAC;qBAAM,CAAC;oBACN,uDAAuD;oBACvD,IAAA,gBAAI,EAAC,gEAAgE,CAAC,CAAC;oBACvE,IAAI,YAAY,CAAC,UAAU,IAAI,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,UAAU,CAAC,EAAE,CAAC;wBACtE,OAAO,YAAY,CAAC,UAAU,CAAC;oBACjC,CAAC;gBACH,CAAC;YACH,CAAC;YAAC,OAAO,UAAU,EAAE,CAAC;gBACpB,IAAA,gBAAI,EAAC,gCAAgC,UAAU,EAAE,CAAC,CAAC;gBACnD,6CAA6C;gBAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;gBAChD,IAAI,SAAS,EAAE,CAAC;oBACd,IAAI,CAAC;wBACH,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;wBAC/C,IAAI,aAAa,CAAC,UAAU,IAAI,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,UAAU,CAAC,EAAE,CAAC;4BACxE,OAAO,aAAa,CAAC,UAAU,CAAC;wBAClC,CAAC;oBACH,CAAC;oBAAC,OAAO,CAAC,EAAE,CAAC;wBACX,IAAA,iBAAK,EAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;oBACtD,CAAC;gBACH,CAAC;YACH,CAAC;YAED,mDAAmD;YACnD,IAAA,iBAAK,EAAC,gDAAgD,CAAC,CAAC;YACxD,OAAO,EAAE,CAAC;QACZ,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,IAAA,iBAAK,EAAC,6BAA6B,CAAC,EAAE,CAAC,CAAC;YACxC,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC;IAED;;OAEG;IACW,aAAa,CAAC,YAA+B;;YACzD,IAAA,eAAG,EAAC,iCAAiC,CAAC,CAAC;YAEvC,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACrD,IAAA,gBAAI,EAAC,2BAA2B,CAAC,CAAC;gBAClC,OAAO,EAAE,CAAC;YACZ,CAAC;YAED,sBAAsB;YACtB,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC;YAE7F,MAAM,MAAM,GAAG,sCAAyB,CAAC,gBAAgB,CACvD,IAAI,CAAC,UAAU,EACf,eAAe,CAChB,CAAC;YAEF,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAEpD,qBAAqB;gBACrB,IAAI,CAAC;oBACH,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;oBAC1C,MAAM,MAAM,GAAG,2BAAc,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;oBAEtD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;wBACnB,OAAO,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;oBAC9B,CAAC;yBAAM,CAAC;wBACN,uDAAuD;wBACvD,IAAA,gBAAI,EAAC,gEAAgE,CAAC,CAAC;wBACvE,IAAI,YAAY,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,QAAQ,CAAC,EAAE,CAAC;4BAClE,OAAO,YAAY,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC;gCAC5C,OAAO,EAAE,CAAC,CAAC,OAAO;gCAClB,MAAM,EAAE,CAAC,CAAC,MAAM;6BACjB,CAAC,CAAC,CAAC;wBACN,CAAC;oBACH,CAAC;gBACH,CAAC;gBAAC,OAAO,UAAU,EAAE,CAAC;oBACpB,IAAA,gBAAI,EAAC,gCAAgC,UAAU,EAAE,CAAC,CAAC;oBACnD,6CAA6C;oBAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;oBAChD,IAAI,SAAS,EAAE,CAAC;wBACd,IAAI,CAAC;4BACH,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;4BAC/C,IAAI,aAAa,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,QAAQ,CAAC,EAAE,CAAC;gCACpE,OAAO,aAAa,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC;oCAC7C,OAAO,EAAE,CAAC,CAAC,OAAO;oCAClB,MAAM,EAAE,CAAC,CAAC,MAAM;iCACjB,CAAC,CAAC,CAAC;4BACN,CAAC;wBACH,CAAC;wBAAC,OAAO,CAAC,EAAE,CAAC;4BACX,IAAA,iBAAK,EAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;wBACtD,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,mDAAmD;gBACnD,IAAA,iBAAK,EAAC,8CAA8C,CAAC,CAAC;gBACtD,OAAO,EAAE,CAAC;YACZ,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,IAAA,iBAAK,EAAC,2BAA2B,CAAC,EAAE,CAAC,CAAC;gBACtC,OAAO,EAAE,CAAC;YACZ,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACK,YAAY,CAAC,YAA+B;QAClD,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACrD,IAAA,gBAAI,EAAC,2BAA2B,CAAC,CAAC;YAClC,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,sBAAsB;QACtB,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC;QAE7F,MAAM,MAAM,GAAG,sCAAyB,CAAC,gBAAgB,CACvD,IAAI,CAAC,UAAU,EACf,eAAe,CAChB,CAAC;QAEF,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YAE7C,qBAAqB;YACrB,IAAI,CAAC;gBACH,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;gBAC1C,MAAM,MAAM,GAAG,2BAAc,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;gBAEtD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;oBACnB,OAAO,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;gBAC9B,CAAC;qBAAM,CAAC;oBACN,uDAAuD;oBACvD,IAAA,gBAAI,EAAC,gEAAgE,CAAC,CAAC;oBACvE,IAAI,YAAY,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,QAAQ,CAAC,EAAE,CAAC;wBAClE,OAAO,YAAY,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC;4BAC5C,OAAO,EAAE,CAAC,CAAC,OAAO;4BAClB,MAAM,EAAE,CAAC,CAAC,MAAM;yBACjB,CAAC,CAAC,CAAC;oBACN,CAAC;gBACH,CAAC;YACH,CAAC;YAAC,OAAO,UAAU,EAAE,CAAC;gBACpB,IAAA,gBAAI,EAAC,gCAAgC,UAAU,EAAE,CAAC,CAAC;gBACnD,6CAA6C;gBAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;gBAChD,IAAI,SAAS,EAAE,CAAC;oBACd,IAAI,CAAC;wBACH,MAAM,aAAa,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC;wBAC/C,IAAI,aAAa,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,QAAQ,CAAC,EAAE,CAAC;4BACpE,OAAO,aAAa,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC;gCAC7C,OAAO,EAAE,CAAC,CAAC,OAAO;gCAClB,MAAM,EAAE,CAAC,CAAC,MAAM;6BACjB,CAAC,CAAC,CAAC;wBACN,CAAC;oBACH,CAAC;oBAAC,OAAO,CAAC,EAAE,CAAC;wBACX,IAAA,iBAAK,EAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;oBACtD,CAAC;gBACH,CAAC;YACH,CAAC;YAED,mDAAmD;YACnD,IAAA,iBAAK,EAAC,8CAA8C,CAAC,CAAC;YACtD,OAAO,EAAE,CAAC;QACZ,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,IAAA,iBAAK,EAAC,2BAA2B,CAAC,EAAE,CAAC,CAAC;YACtC,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC;IAED;;OAEG;IACW,WAAW;;;YACvB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;gBACzB,OAAO,SAAS,CAAC;YACnB,CAAC;YAED,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACjD,OAAO,SAAS,CAAC;YACnB,CAAC;YAED,IAAI,CAAC;gBACH,+CAA+C;gBAC/C,MAAM,mBAAmB,GAAuB,EAAE,CAAC;gBACnD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,UAAW,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;oBACjD,IAAI,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,IAAI,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,IAAI,EAAE,CAAC;wBAChF,mBAAmB,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,UAAW,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;oBAC3E,CAAC;gBACH,CAAC;gBAED,IAAI,mBAAmB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACrC,OAAO,uFAAuF,CAAC;gBACjG,CAAC;gBAED,kBAAkB;gBAClB,MAAM,MAAM,GAAG,sCAAyB,CAAC,cAAc,CACrD,mBAAmB,EACnB,CAAA,MAAA,IAAI,CAAC,KAAK,0CAAE,QAAQ,EAAE,KAAI,GAAG,CAC9B,CAAC;gBAEF,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBACtD,MAAM,YAAY,GAAG,yBAAY,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;gBAEpE,IAAI,CAAC,YAAY,CAAC,OAAO,EAAE,CAAC;oBAC1B,IAAA,iBAAK,EAAC,2BAA2B,YAAY,CAAC,KAAK,EAAE,CAAC,CAAC;oBACvD,OAAO,SAAS,CAAC;gBACnB,CAAC;gBAED,OAAO,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC;YAClC,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,IAAA,iBAAK,EAAC,yBAAyB,GAAG,EAAE,CAAC,CAAC;gBACtC,OAAO,SAAS,CAAC;YACnB,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACK,UAAU;;QAChB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;YACzB,OAAO,SAAS,CAAC;QACnB,CAAC;QAED,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACjD,OAAO,SAAS,CAAC;QACnB,CAAC;QAED,IAAI,CAAC;YACH,+CAA+C;YAC/C,MAAM,mBAAmB,GAAuB,EAAE,CAAC;YACnD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,UAAW,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACjD,IAAI,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,IAAI,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,IAAI,EAAE,CAAC;oBAChF,mBAAmB,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,UAAW,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;gBAC3E,CAAC;YACH,CAAC;YAED,IAAI,mBAAmB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACrC,OAAO,uFAAuF,CAAC;YACjG,CAAC;YAED,kBAAkB;YAClB,MAAM,MAAM,GAAG,sCAAyB,CAAC,cAAc,CACrD,mBAAmB,EACnB,CAAA,MAAA,IAAI,CAAC,KAAK,0CAAE,QAAQ,EAAE,KAAI,GAAG,CAC9B,CAAC;YAEF,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YAC/C,MAAM,YAAY,GAAG,yBAAY,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;YAEpE,IAAI,CAAC,YAAY,CAAC,OAAO,EAAE,CAAC;gBAC1B,IAAA,iBAAK,EAAC,2BAA2B,YAAY,CAAC,KAAK,EAAE,CAAC,CAAC;gBACvD,OAAO,SAAS,CAAC;YACnB,CAAC;YAED,OAAO,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC;QAClC,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAA,iBAAK,EAAC,yBAAyB,GAAG,EAAE,CAAC,CAAC;YACtC,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,aAAa;QACnB,IAAA,eAAG,EAAC,iBAAiB,CAAC,CAAC;QAEvB,gFAAgF;QAChF,qFAAqF;QACrF,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACrD,OAAO,CAAC,CAAC;QACX,CAAC;QAED,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACjD,OAAO,CAAC,CAAC;QACX,CAAC;QAED,IAAI,YAAY,GAAG,CAAC,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YACpC,IAAI,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,KAAK,KAAK,EAAE,CAAC;gBACnD,YAAY,EAAE,CAAC;YACjB,CAAC;QACH,CAAC;QAED,MAAM,KAAK,GAAG,YAAY,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC;QAElD,wDAAwD;QACxD,OAAO,IAAI,CAAC,WAAW,IAAI,KAAK,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;IAChE,CAAC;IAED;;OAEG;IACO,mBAAmB,CAAC,OAAgB;QAC5C,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;YACxC,IAAI,KAAK,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;YAClE,CAAC;iBAAM,IAAI,KAAK,KAAK,cAAc,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC;gBAC7D,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;YACzE,CAAC;iBAAM,IAAI,KAAK,KAAK,gBAAgB,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,CAAC;gBACjE,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAC;YAC3E,CAAC;iBAAM,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;gBACxF,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;YACvF,CAAC;iBAAM,IAAI,KAAK,KAAK,kBAAkB,IAAI,CAAC,CAAC,OAAO,CAAC,gBAAgB,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,EAAE,CAAC;gBACnH,MAAM,IAAI,KAAK,CAAC,4EAA4E,CAAC,CAAC;YAChG,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACK,kBAAkB;QACxB,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,MAAM,KAAK,GAAG;YACZ,gBAAgB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,UAAU,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE;YAC1D,cAAc,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE;YACtD,UAAU,IAAI,CAAC,KAAK,aAAa,IAAI,CAAC,MAAM,EAAE;SAC/C,CAAC;QAEF,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,OAAgB;QAC/B,IAAA,gBAAI,EAAC,sCAAsC,CAAC,CAAC;QAE7C,IAAI,CAAC;YACH,4BAA4B;YAC5B,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC;YAElC,kBAAkB;YAClB,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;gBACpB,MAAM,IAAI,KAAK,CAAC,2FAA2F,CAAC,CAAC;YAC/G,CAAC;YAED,oBAAoB;YACpB,IAAI,YAAY,GAAG,CAAC,CAAC;YACrB,IAAI,gBAAgB,GAAG,CAAC,CAAC;YAEzB,kCAAkC;YAClC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,cAAe,CAAC,CAAC;YAC/D,YAAY,IAAI,GAAG,CAAC,CAAC,2CAA2C;YAChE,gBAAgB,IAAI,GAAG,CAAC,CAAC,6CAA6C;YAEtE,gCAAgC;YAChC,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,YAAY,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC;YACzD,YAAY,IAAI,GAAG,CAAC,CAAC,yCAAyC;YAC9D,gBAAgB,IAAI,GAAG,CAAC,CAAC,2CAA2C;YAEpE,gBAAgB;YAChB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;YAElC,uBAAuB;YACvB,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;gBACxB,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;gBAChC,YAAY,IAAI,GAAG,CAAC,CAAC,uCAAuC;gBAC5D,gBAAgB,IAAI,GAAG,CAAC,CAAC,yCAAyC;YACpE,CAAC;YAED,4BAA4B;YAC5B,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,oBAAoB,CAC9C,IAAI,CAAC,gBAAgB,IAAI,eAAe,EACxC,YAAY,EACZ,gBAAgB,CACjB,CAAC;YAEF,mBAAmB;YACnB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,SAAS,CAAC;YAE5C,gCAAgC;YAChC,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;gBACtB,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAChD,CAAC;YAED,IAAA,gBAAI,EAAC,iCAAiC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;YAEpD,mDAAmD;YACnD,OAAO;gBACL,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,KAAK;gBAC9B,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,CAAC;gBACtB,MAAM,EAAE,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI;gBACtD,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,KAAK;gBACtC,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAC/C,KAAK,EAAE,IAAI;gBACX,eAAe,EAAE,IAAI,CAAC,eAAe;gBACrC,YAAY,EAAE,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI;gBAC1D,mBAAmB,EAAE,IAAI,CAAC,mBAAmB,IAAI,EAAE;aACpD,CAAC;QACJ,CAAC;QAAC,OAAO,KAAU,EAAE,CAAC;YACpB,gBAAgB;YAChB,MAAM,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC5E,IAAI,CAAC,KAAK,GAAG,YAAY,CAAC;YAC1B,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC;YAErB,OAAO;gBACL,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,KAAK;gBACd,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,yBAAyB,YAAY,EAAE;gBAC/C,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,KAAK;gBACtC,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAC/C,KAAK,EAAE,YAAY;gBACnB,eAAe,EAAE,IAAI;gBACrB,YAAY,EAAE,IAAI;gBAClB,mBAAmB,EAAE,IAAI,CAAC,mBAAmB,IAAI,EAAE;aACpD,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;;OAGG;IACG,YAAY,CAAC,OAAgB;;YACjC,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC;gBACrB,OAAO,IAAI,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC;YACxC,CAAC;YAED,IAAA,gBAAI,EAAC,uCAAuC,CAAC,CAAC;YAE9C,IAAI,CAAC;gBACH,4BAA4B;gBAC5B,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC;gBAElC,oBAAoB;gBACpB,IAAI,YAAY,GAAG,CAAC,CAAC;gBACrB,IAAI,gBAAgB,GAAG,CAAC,CAAC;gBAEzB,kBAAkB;gBAClB,IAAI,CAAC,UAAU,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,cAAe,CAAC,CAAC;gBACtE,YAAY,IAAI,GAAG,CAAC,CAAC,2CAA2C;gBAChE,gBAAgB,IAAI,GAAG,CAAC,CAAC,6CAA6C;gBAEtE,IAAI,CAAC,QAAQ,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC;gBAChE,YAAY,IAAI,GAAG,CAAC,CAAC,yCAAyC;gBAC9D,gBAAgB,IAAI,GAAG,CAAC,CAAC,2CAA2C;gBAEpE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;gBAElC,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;oBACxB,IAAI,CAAC,MAAM,GAAG,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;oBACvC,YAAY,IAAI,GAAG,CAAC,CAAC,uCAAuC;oBAC5D,gBAAgB,IAAI,GAAG,CAAC,CAAC,yCAAyC;gBACpE,CAAC;gBAED,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;gBAEpC,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;oBACtB,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,kBAAkB,EAAE,CAAC;gBAChD,CAAC;gBAED,4BAA4B;gBAC5B,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,oBAAoB,CAC9C,IAAI,CAAC,gBAAgB,IAAI,eAAe,EACxC,YAAY,EACZ,gBAAgB,CACjB,CAAC;gBAEF,IAAA,gBAAI,EAAC,iCAAiC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;gBAEpD,mDAAmD;gBACnD,OAAO;oBACL,IAAI,EAAE,IAAI,CAAC,IAAI;oBACf,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,KAAK;oBAC9B,KAAK,EAAE,IAAI,CAAC,KAAK,IAAI,CAAC;oBACtB,MAAM,EAAE,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI;oBACtD,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,KAAK;oBACtC,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;oBAC/C,KAAK,EAAE,IAAI;oBACX,eAAe,EAAE,IAAI,CAAC,eAAe;oBACrC,YAAY,EAAE,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI;oBAC1D,mBAAmB,EAAE,IAAI,CAAC,mBAAmB,IAAI,EAAE;iBACpD,CAAC;YACJ,CAAC;YAAC,OAAO,KAAU,EAAE,CAAC;gBACpB,gBAAgB;gBAChB,MAAM,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;gBAC5E,IAAI,CAAC,KAAK,GAAG,YAAY,CAAC;gBAC1B,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC;gBAErB,OAAO;oBACL,IAAI,EAAE,IAAI,CAAC,IAAI;oBACf,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,OAAO,EAAE,KAAK;oBACd,KAAK,EAAE,CAAC;oBACR,MAAM,EAAE,yBAAyB,YAAY,EAAE;oBAC/C,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,KAAK;oBACtC,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;oBAC/C,KAAK,EAAE,YAAY;oBACnB,eAAe,EAAE,IAAI;oBACrB,YAAY,EAAE,IAAI;oBAClB,mBAAmB,EAAE,IAAI,CAAC,mBAAmB,IAAI,EAAE;iBACpD,CAAC;YACJ,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACH,IAAI,IAAI;QACN,OAAO,oBAAoB,CAAC;IAC9B,CAAC;CACF;AAtnBD,0DAsnBC"}
|
|
@@ -1,19 +0,0 @@
|
|
|
1
|
-
"use strict";
|
|
2
|
-
var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
|
|
3
|
-
if (k2 === undefined) k2 = k;
|
|
4
|
-
var desc = Object.getOwnPropertyDescriptor(m, k);
|
|
5
|
-
if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
|
|
6
|
-
desc = { enumerable: true, get: function() { return m[k]; } };
|
|
7
|
-
}
|
|
8
|
-
Object.defineProperty(o, k2, desc);
|
|
9
|
-
}) : (function(o, m, k, k2) {
|
|
10
|
-
if (k2 === undefined) k2 = k;
|
|
11
|
-
o[k2] = m[k];
|
|
12
|
-
}));
|
|
13
|
-
var __exportStar = (this && this.__exportStar) || function(m, exports) {
|
|
14
|
-
for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
|
|
15
|
-
};
|
|
16
|
-
Object.defineProperty(exports, "__esModule", { value: true });
|
|
17
|
-
__exportStar(require("./answer-correctness.js"), exports);
|
|
18
|
-
__exportStar(require("./prompts.js"), exports);
|
|
19
|
-
//# sourceMappingURL=index.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../../src/scorers/metrics/answer-correctness/index.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;AAAA,0DAAwC;AACxC,+CAA6B"}
|