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
|
@@ -1,208 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Logger utilities and result printing for the JudgEval TypeScript SDK
|
|
3
|
-
*/
|
|
4
|
-
// Remove fs and path imports if no longer needed
|
|
5
|
-
// import * as fs from 'fs';
|
|
6
|
-
// import * as path from 'path';
|
|
7
|
-
import logger from './logger-instance.js'; // Import the configured winston logger
|
|
8
|
-
// Track current example info (Keep for potential context integration)
|
|
9
|
-
let currentExampleId = null;
|
|
10
|
-
let currentTimestamp = null;
|
|
11
|
-
/**
|
|
12
|
-
* Log a debug message
|
|
13
|
-
*/
|
|
14
|
-
export function debug(message, meta) {
|
|
15
|
-
// Prioritize meta.exampleId over currentExampleId
|
|
16
|
-
const exampleId = (meta === null || meta === void 0 ? void 0 : meta.exampleId) !== undefined ? meta.exampleId : currentExampleId;
|
|
17
|
-
logger.debug(message, Object.assign(Object.assign({}, meta), { exampleId, timestamp: currentTimestamp }));
|
|
18
|
-
}
|
|
19
|
-
/**
|
|
20
|
-
* Log an info message (alias for info)
|
|
21
|
-
*/
|
|
22
|
-
export function log(message, meta) {
|
|
23
|
-
// Prioritize meta.exampleId over currentExampleId
|
|
24
|
-
const exampleId = (meta === null || meta === void 0 ? void 0 : meta.exampleId) !== undefined ? meta.exampleId : currentExampleId;
|
|
25
|
-
logger.info(message, Object.assign(Object.assign({}, meta), { exampleId, timestamp: currentTimestamp }));
|
|
26
|
-
}
|
|
27
|
-
/**
|
|
28
|
-
* Log an info message
|
|
29
|
-
*/
|
|
30
|
-
export function info(message, meta) {
|
|
31
|
-
// Prioritize meta.exampleId over currentExampleId
|
|
32
|
-
const exampleId = (meta === null || meta === void 0 ? void 0 : meta.exampleId) !== undefined ? meta.exampleId : currentExampleId;
|
|
33
|
-
logger.info(message, Object.assign(Object.assign({}, meta), { exampleId, timestamp: currentTimestamp }));
|
|
34
|
-
}
|
|
35
|
-
/**
|
|
36
|
-
* Log a warning message
|
|
37
|
-
*/
|
|
38
|
-
export function warning(message, meta) {
|
|
39
|
-
// Prioritize meta.exampleId over currentExampleId
|
|
40
|
-
const exampleId = (meta === null || meta === void 0 ? void 0 : meta.exampleId) !== undefined ? meta.exampleId : currentExampleId;
|
|
41
|
-
logger.warn(message, Object.assign(Object.assign({}, meta), { exampleId, timestamp: currentTimestamp }));
|
|
42
|
-
}
|
|
43
|
-
/**
|
|
44
|
-
* Alias for warning
|
|
45
|
-
*/
|
|
46
|
-
export function warn(message, meta) {
|
|
47
|
-
warning(message, meta);
|
|
48
|
-
}
|
|
49
|
-
/**
|
|
50
|
-
* Log an error message
|
|
51
|
-
*/
|
|
52
|
-
export function error(message, meta) {
|
|
53
|
-
// Prioritize meta.exampleId over currentExampleId
|
|
54
|
-
const exampleId = (meta === null || meta === void 0 ? void 0 : meta.exampleId) !== undefined ? meta.exampleId : currentExampleId;
|
|
55
|
-
logger.error(message, Object.assign(Object.assign({}, meta), { exampleId, timestamp: currentTimestamp }));
|
|
56
|
-
}
|
|
57
|
-
/**
|
|
58
|
-
* Set the current example context for logging (Keep for potential context integration)
|
|
59
|
-
*/
|
|
60
|
-
export function setExampleContext(exampleId, timestamp) {
|
|
61
|
-
currentExampleId = exampleId;
|
|
62
|
-
currentTimestamp = timestamp;
|
|
63
|
-
}
|
|
64
|
-
/**
|
|
65
|
-
* Clear the current example context (Keep for potential context integration)
|
|
66
|
-
*/
|
|
67
|
-
export function clearExampleContext() {
|
|
68
|
-
currentExampleId = null;
|
|
69
|
-
currentTimestamp = null;
|
|
70
|
-
}
|
|
71
|
-
/**
|
|
72
|
-
* Create a context for example-specific logging (Keep for potential context integration)
|
|
73
|
-
*/
|
|
74
|
-
export function withExampleContext(exampleId, timestamp, fn) {
|
|
75
|
-
setExampleContext(exampleId, timestamp);
|
|
76
|
-
try {
|
|
77
|
-
return fn();
|
|
78
|
-
}
|
|
79
|
-
finally {
|
|
80
|
-
clearExampleContext();
|
|
81
|
-
}
|
|
82
|
-
}
|
|
83
|
-
/**
|
|
84
|
-
* Format evaluation results for display
|
|
85
|
-
* This matches the Python SDK's output format with clickable links
|
|
86
|
-
*/
|
|
87
|
-
export function formatEvaluationResults(results, projectName, evalName) {
|
|
88
|
-
var _a;
|
|
89
|
-
let output = '';
|
|
90
|
-
// Print summary information
|
|
91
|
-
if (results.length > 0) {
|
|
92
|
-
output += `\n=== Evaluation Results (${results.length} examples) ===\n\n`;
|
|
93
|
-
// Calculate success rate
|
|
94
|
-
const successfulExamples = results.filter(r => { var _a, _b; return r.success || ((_b = (_a = r.scorersData) === null || _a === void 0 ? void 0 : _a.every((s) => s.success)) !== null && _b !== void 0 ? _b : false); }).length;
|
|
95
|
-
const successRate = (successfulExamples / results.length) * 100;
|
|
96
|
-
output += `Success Rate: ${successRate.toFixed(2)}% (${successfulExamples}/${results.length})\n\n`;
|
|
97
|
-
// Print failures if any
|
|
98
|
-
const failures = results.filter(r => { var _a, _b; return !r.success || ((_b = (_a = r.scorersData) === null || _a === void 0 ? void 0 : _a.some((s) => !s.success)) !== null && _b !== void 0 ? _b : false); });
|
|
99
|
-
if (failures.length > 0) {
|
|
100
|
-
output += `Failures (${failures.length}):\n`;
|
|
101
|
-
for (const [index, failure] of failures.entries()) {
|
|
102
|
-
output += `\nExample ${index + 1}:\n`;
|
|
103
|
-
output += `Input: ${((_a = failure.example) === null || _a === void 0 ? void 0 : _a.input) || 'N/A'}\n`;
|
|
104
|
-
if (failure.scorersData) {
|
|
105
|
-
output += 'Scorer Failures:\n';
|
|
106
|
-
for (const scorer of failure.scorersData) {
|
|
107
|
-
if (!scorer.success) {
|
|
108
|
-
output += ` - ${scorer.name}: ${scorer.error || 'Unknown error'}\n`;
|
|
109
|
-
}
|
|
110
|
-
}
|
|
111
|
-
}
|
|
112
|
-
else if (!failure.success) {
|
|
113
|
-
output += `Error: ${failure.error || 'Unknown error'}\n`;
|
|
114
|
-
}
|
|
115
|
-
}
|
|
116
|
-
}
|
|
117
|
-
}
|
|
118
|
-
return output;
|
|
119
|
-
}
|
|
120
|
-
/**
|
|
121
|
-
* Print evaluation results to the console
|
|
122
|
-
* This matches the Python SDK's output format exactly
|
|
123
|
-
*/
|
|
124
|
-
export function printResults(results, projectName, evalName) {
|
|
125
|
-
// Always print a URL if projectName and evalName are provided
|
|
126
|
-
if (projectName && evalName) {
|
|
127
|
-
const baseUrl = 'https://app.judgmentlabs.ai/app/experiment';
|
|
128
|
-
const urlParams = `?project_name=${projectName}&eval_run_name=${evalName}`;
|
|
129
|
-
const resultsUrl = `${baseUrl}${urlParams}`;
|
|
130
|
-
// Print the URL
|
|
131
|
-
console.log(`\n🔍 View results: ${resultsUrl}\n`);
|
|
132
|
-
}
|
|
133
|
-
// Format the results - only includes failure details
|
|
134
|
-
const formattedResults = formatEvaluationResults(results, projectName, evalName);
|
|
135
|
-
// Print the results to the console directly
|
|
136
|
-
if (formattedResults) {
|
|
137
|
-
console.log(formattedResults);
|
|
138
|
-
}
|
|
139
|
-
// Print raw results in the same format as Python SDK
|
|
140
|
-
console.log(JSON.stringify(results, null, 2));
|
|
141
|
-
}
|
|
142
|
-
/**
|
|
143
|
-
* Simplified print function for results - matches Python SDK's print(results) behavior
|
|
144
|
-
* This is the preferred way to print results
|
|
145
|
-
*/
|
|
146
|
-
export function print(data) {
|
|
147
|
-
if (Array.isArray(data)) {
|
|
148
|
-
// Handle array of results (evaluation results)
|
|
149
|
-
let projectName, evalName;
|
|
150
|
-
// Try to extract project name and eval name from the first result
|
|
151
|
-
if (data.length > 0 && data[0].metadata) {
|
|
152
|
-
projectName = data[0].metadata.project_name;
|
|
153
|
-
evalName = data[0].metadata.eval_name;
|
|
154
|
-
}
|
|
155
|
-
printResults(data, projectName, evalName);
|
|
156
|
-
}
|
|
157
|
-
else if (data && typeof data === 'object' && data.traceId) {
|
|
158
|
-
// Handle trace object
|
|
159
|
-
console.log(`\n--- Trace: ${data.name || 'Unnamed'} (ID: ${data.traceId}) ---`);
|
|
160
|
-
if (data.projectName) {
|
|
161
|
-
const traceUrl = `https://app.judgmentlabs.ai/app/monitor?project_name=${data.projectName}&trace_id=${data.traceId}&trace_name=${data.name || 'trace'}&show_trace=true`;
|
|
162
|
-
console.log(`\n🔍 View trace: ${traceUrl}\n`);
|
|
163
|
-
}
|
|
164
|
-
console.log(JSON.stringify(data, null, 2));
|
|
165
|
-
}
|
|
166
|
-
else if (data && typeof data === 'object' && data.title === "Workflow Analysis Results") {
|
|
167
|
-
// Handle workflow analysis results
|
|
168
|
-
console.log(`\n=== ${data.title} ===\n`);
|
|
169
|
-
// Print scorer performance
|
|
170
|
-
console.log('Scorer Performance Summary:');
|
|
171
|
-
console.log('----------------------------');
|
|
172
|
-
if (Array.isArray(data.scorerPerformance)) {
|
|
173
|
-
data.scorerPerformance.forEach((scorer) => {
|
|
174
|
-
console.log(`${scorer.name.padEnd(30)} ${scorer.score.toFixed(2)} (${scorer.rating})`);
|
|
175
|
-
});
|
|
176
|
-
}
|
|
177
|
-
// Print areas for improvement
|
|
178
|
-
if (Array.isArray(data.areasForImprovement) && data.areasForImprovement.length > 0) {
|
|
179
|
-
console.log('\nAreas for Improvement:');
|
|
180
|
-
console.log('----------------------');
|
|
181
|
-
data.areasForImprovement.forEach((area, index) => {
|
|
182
|
-
console.log(`${index + 1}. ${area}`);
|
|
183
|
-
});
|
|
184
|
-
}
|
|
185
|
-
// Print strengths
|
|
186
|
-
if (Array.isArray(data.strengths) && data.strengths.length > 0) {
|
|
187
|
-
console.log('\nStrengths:');
|
|
188
|
-
console.log('----------');
|
|
189
|
-
data.strengths.forEach((strength, index) => {
|
|
190
|
-
console.log(`${index + 1}. ${strength}`);
|
|
191
|
-
});
|
|
192
|
-
}
|
|
193
|
-
}
|
|
194
|
-
else if (data && typeof data === 'object' && data.title && data.recommendations) {
|
|
195
|
-
// Handle recommendations
|
|
196
|
-
console.log(`\n=== ${data.title} ===`);
|
|
197
|
-
if (Array.isArray(data.recommendations)) {
|
|
198
|
-
data.recommendations.forEach((rec, index) => {
|
|
199
|
-
console.log(`${index + 1}. ${rec}`);
|
|
200
|
-
});
|
|
201
|
-
}
|
|
202
|
-
}
|
|
203
|
-
else {
|
|
204
|
-
// Handle any other object
|
|
205
|
-
console.log(JSON.stringify(data, null, 2));
|
|
206
|
-
}
|
|
207
|
-
}
|
|
208
|
-
//# sourceMappingURL=logger.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"logger.js","sourceRoot":"","sources":["../../../src/common/logger.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,iDAAiD;AACjD,4BAA4B;AAC5B,gCAAgC;AAChC,OAAO,MAAM,MAAM,sBAAsB,CAAC,CAAC,uCAAuC;AAElF,sEAAsE;AACtE,IAAI,gBAAgB,GAAkB,IAAI,CAAC;AAC3C,IAAI,gBAAgB,GAAkB,IAAI,CAAC;AAE3C;;GAEG;AACH,MAAM,UAAU,KAAK,CAAC,OAAe,EAAE,IAA0B;IAC/D,kDAAkD;IAClD,MAAM,SAAS,GAAG,CAAA,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,SAAS,MAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACpF,MAAM,CAAC,KAAK,CAAC,OAAO,kCAAO,IAAI,KAAE,SAAS,EAAE,SAAS,EAAE,gBAAgB,IAAG,CAAC;AAC7E,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,GAAG,CAAC,OAAe,EAAE,IAA0B;IAC7D,kDAAkD;IAClD,MAAM,SAAS,GAAG,CAAA,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,SAAS,MAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACpF,MAAM,CAAC,IAAI,CAAC,OAAO,kCAAO,IAAI,KAAE,SAAS,EAAE,SAAS,EAAE,gBAAgB,IAAG,CAAC;AAC5E,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,IAAI,CAAC,OAAe,EAAE,IAA0B;IAC9D,kDAAkD;IAClD,MAAM,SAAS,GAAG,CAAA,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,SAAS,MAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACpF,MAAM,CAAC,IAAI,CAAC,OAAO,kCAAO,IAAI,KAAE,SAAS,EAAE,SAAS,EAAE,gBAAgB,IAAG,CAAC;AAC5E,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,OAAO,CAAC,OAAe,EAAE,IAA0B;IACjE,kDAAkD;IAClD,MAAM,SAAS,GAAG,CAAA,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,SAAS,MAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACpF,MAAM,CAAC,IAAI,CAAC,OAAO,kCAAO,IAAI,KAAE,SAAS,EAAE,SAAS,EAAE,gBAAgB,IAAG,CAAC;AAC5E,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,IAAI,CAAC,OAAe,EAAE,IAA0B;IAC9D,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;AACzB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,KAAK,CAAC,OAAe,EAAE,IAA0B;IAC/D,kDAAkD;IAClD,MAAM,SAAS,GAAG,CAAA,IAAI,aAAJ,IAAI,uBAAJ,IAAI,CAAE,SAAS,MAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,gBAAgB,CAAC;IACpF,MAAM,CAAC,KAAK,CAAC,OAAO,kCAAO,IAAI,KAAE,SAAS,EAAE,SAAS,EAAE,gBAAgB,IAAG,CAAC;AAC7E,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB,CAAC,SAAiB,EAAE,SAAiB;IACpE,gBAAgB,GAAG,SAAS,CAAC;IAC7B,gBAAgB,GAAG,SAAS,CAAC;AAC/B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IACjC,gBAAgB,GAAG,IAAI,CAAC;IACxB,gBAAgB,GAAG,IAAI,CAAC;AAC1B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,kBAAkB,CAAI,SAAiB,EAAE,SAAiB,EAAE,EAAW;IACrF,iBAAiB,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;IACxC,IAAI,CAAC;QACH,OAAO,EAAE,EAAE,CAAC;IACd,CAAC;YAAS,CAAC;QACT,mBAAmB,EAAE,CAAC;IACxB,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,uBAAuB,CAAC,OAAc,EAAE,WAAoB,EAAE,QAAiB;;IAC7F,IAAI,MAAM,GAAG,EAAE,CAAC;IAEhB,4BAA4B;IAC5B,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,6BAA6B,OAAO,CAAC,MAAM,oBAAoB,CAAC;QAE1E,yBAAyB;QACzB,MAAM,kBAAkB,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,eAC5C,OAAA,CAAC,CAAC,OAAO,IAAI,CAAC,MAAA,MAAA,CAAC,CAAC,WAAW,0CAAE,KAAK,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,mCAAI,KAAK,CAAC,CAAA,EAAA,CACpE,CAAC,MAAM,CAAC;QAET,MAAM,WAAW,GAAG,CAAC,kBAAkB,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC;QAChE,MAAM,IAAI,iBAAiB,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,kBAAkB,IAAI,OAAO,CAAC,MAAM,OAAO,CAAC;QAEnG,wBAAwB;QACxB,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,eAClC,OAAA,CAAC,CAAC,CAAC,OAAO,IAAI,CAAC,MAAA,MAAA,CAAC,CAAC,WAAW,0CAAE,IAAI,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,mCAAI,KAAK,CAAC,CAAA,EAAA,CACrE,CAAC;QAEF,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,aAAa,QAAQ,CAAC,MAAM,MAAM,CAAC;YAE7C,KAAK,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,IAAI,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC;gBAClD,MAAM,IAAI,aAAa,KAAK,GAAG,CAAC,KAAK,CAAC;gBACtC,MAAM,IAAI,UAAU,CAAA,MAAA,OAAO,CAAC,OAAO,0CAAE,KAAK,KAAI,KAAK,IAAI,CAAC;gBAExD,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC;oBACxB,MAAM,IAAI,oBAAoB,CAAC;oBAE/B,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC;wBACzC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;4BACpB,MAAM,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,KAAK,IAAI,eAAe,IAAI,CAAC;wBACvE,CAAC;oBACH,CAAC;gBACH,CAAC;qBAAM,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;oBAC5B,MAAM,IAAI,UAAU,OAAO,CAAC,KAAK,IAAI,eAAe,IAAI,CAAC;gBAC3D,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,YAAY,CAAC,OAAc,EAAE,WAAoB,EAAE,QAAiB;IAClF,8DAA8D;IAC9D,IAAI,WAAW,IAAI,QAAQ,EAAE,CAAC;QAC5B,MAAM,OAAO,GAAG,4CAA4C,CAAC;QAC7D,MAAM,SAAS,GAAG,iBAAiB,WAAW,kBAAkB,QAAQ,EAAE,CAAC;QAC3E,MAAM,UAAU,GAAG,GAAG,OAAO,GAAG,SAAS,EAAE,CAAC;QAE5C,gBAAgB;QAChB,OAAO,CAAC,GAAG,CAAC,sBAAsB,UAAU,IAAI,CAAC,CAAC;IACpD,CAAC;IAED,qDAAqD;IACrD,MAAM,gBAAgB,GAAG,uBAAuB,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;IAEjF,4CAA4C;IAC5C,IAAI,gBAAgB,EAAE,CAAC;QACrB,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;IAChC,CAAC;IAED,qDAAqD;IACrD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;AAChD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,KAAK,CAAC,IAAS;IAC7B,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC;QACxB,+CAA+C;QAC/C,IAAI,WAAW,EAAE,QAAQ,CAAC;QAE1B,kEAAkE;QAClE,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC;YACxC,WAAW,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,YAAY,CAAC;YAC5C,QAAQ,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC;QACxC,CAAC;QAED,YAAY,CAAC,IAAI,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;IAC5C,CAAC;SAAM,IAAI,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;QAC5D,sBAAsB;QACtB,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,CAAC,IAAI,IAAI,SAAS,SAAS,IAAI,CAAC,OAAO,OAAO,CAAC,CAAC;QAChF,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YACrB,MAAM,QAAQ,GAAG,wDAAwD,IAAI,CAAC,WAAW,aAAa,IAAI,CAAC,OAAO,eAAe,IAAI,CAAC,IAAI,IAAI,OAAO,kBAAkB,CAAC;YACxK,OAAO,CAAC,GAAG,CAAC,oBAAoB,QAAQ,IAAI,CAAC,CAAC;QAChD,CAAC;QACD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7C,CAAC;SAAM,IAAI,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,KAAK,KAAK,2BAA2B,EAAE,CAAC;QAC1F,mCAAmC;QACnC,OAAO,CAAC,GAAG,CAAC,SAAS,IAAI,CAAC,KAAK,QAAQ,CAAC,CAAC;QAEzC,2BAA2B;QAC3B,OAAO,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC;QAC3C,OAAO,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC;QAC5C,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,iBAAiB,CAAC,EAAE,CAAC;YAC1C,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,MAAW,EAAE,EAAE;gBAC7C,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;YACzF,CAAC,CAAC,CAAC;QACL,CAAC;QAED,8BAA8B;QAC9B,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,mBAAmB,CAAC,IAAI,IAAI,CAAC,mBAAmB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACnF,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC,CAAC;YACxC,OAAO,CAAC,GAAG,CAAC,wBAAwB,CAAC,CAAC;YACtC,IAAI,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC,IAAY,EAAE,KAAa,EAAE,EAAE;gBAC/D,OAAO,CAAC,GAAG,CAAC,GAAG,KAAK,GAAG,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;YACvC,CAAC,CAAC,CAAC;QACL,CAAC;QAED,kBAAkB;QAClB,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC/D,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;YAC5B,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAC1B,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,QAAgB,EAAE,KAAa,EAAE,EAAE;gBACzD,OAAO,CAAC,GAAG,CAAC,GAAG,KAAK,GAAG,CAAC,KAAK,QAAQ,EAAE,CAAC,CAAC;YAC3C,CAAC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;SAAM,IAAI,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,eAAe,EAAE,CAAC;QAClF,yBAAyB;QACzB,OAAO,CAAC,GAAG,CAAC,SAAS,IAAI,CAAC,KAAK,MAAM,CAAC,CAAC;QAEvC,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC,EAAE,CAAC;YACxC,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC,GAAW,EAAE,KAAa,EAAE,EAAE;gBAC1D,OAAO,CAAC,GAAG,CAAC,GAAG,KAAK,GAAG,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC;YACtC,CAAC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;SAAM,CAAC;QACN,0BAA0B;QAC1B,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC7C,CAAC;AACH,CAAC"}
|
|
@@ -1,91 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Token cost utilities for the JudgEval TypeScript SDK
|
|
3
|
-
* This provides a local implementation of token cost calculation
|
|
4
|
-
* similar to LiteLLM's cost_per_token function in Python
|
|
5
|
-
*/
|
|
6
|
-
import { warn } from './logger.js';
|
|
7
|
-
// Model cost mapping (USD per 1000 tokens)
|
|
8
|
-
// These rates are based on OpenAI's pricing as of April 2025
|
|
9
|
-
// Source: https://openai.com/pricing
|
|
10
|
-
const MODEL_COSTS = {
|
|
11
|
-
// GPT-4 models
|
|
12
|
-
'gpt-4': { prompt: 0.03, completion: 0.06 },
|
|
13
|
-
'gpt-4-32k': { prompt: 0.06, completion: 0.12 },
|
|
14
|
-
'gpt-4-turbo': { prompt: 0.01, completion: 0.03 },
|
|
15
|
-
'gpt-4-turbo-preview': { prompt: 0.01, completion: 0.03 },
|
|
16
|
-
'gpt-4-vision-preview': { prompt: 0.01, completion: 0.03 },
|
|
17
|
-
'gpt-4o': { prompt: 0.005, completion: 0.015 },
|
|
18
|
-
'gpt-4o-mini': { prompt: 0.0015, completion: 0.006 },
|
|
19
|
-
// GPT-3.5 models
|
|
20
|
-
'gpt-3.5-turbo': { prompt: 0.0005, completion: 0.0015 },
|
|
21
|
-
'gpt-3.5-turbo-16k': { prompt: 0.001, completion: 0.002 },
|
|
22
|
-
'gpt-3.5-turbo-instruct': { prompt: 0.0015, completion: 0.002 },
|
|
23
|
-
// Claude models
|
|
24
|
-
'claude-instant-1': { prompt: 0.0008, completion: 0.0024 },
|
|
25
|
-
'claude-2': { prompt: 0.008, completion: 0.024 },
|
|
26
|
-
'claude-2.1': { prompt: 0.008, completion: 0.024 },
|
|
27
|
-
'claude-3-opus-20240229': { prompt: 0.015, completion: 0.075 },
|
|
28
|
-
'claude-3-sonnet-20240229': { prompt: 0.003, completion: 0.015 },
|
|
29
|
-
'claude-3-haiku-20240307': { prompt: 0.00025, completion: 0.00125 },
|
|
30
|
-
// Default fallback
|
|
31
|
-
'default': { prompt: 0.001, completion: 0.002 }
|
|
32
|
-
};
|
|
33
|
-
/**
|
|
34
|
-
* Calculate cost per token for a given model
|
|
35
|
-
* This mimics LiteLLM's cost_per_token function in Python
|
|
36
|
-
*
|
|
37
|
-
* @param model Model name (e.g., 'gpt-3.5-turbo')
|
|
38
|
-
* @param promptTokens Number of prompt tokens (used for total calculation)
|
|
39
|
-
* @param completionTokens Number of completion tokens (used for total calculation)
|
|
40
|
-
* @returns [promptCostPerToken, completionCostPerToken] - Cost per token for prompt and completion
|
|
41
|
-
*/
|
|
42
|
-
export function costPerToken(model, promptTokens = 1, completionTokens = 1) {
|
|
43
|
-
// Find the model in our mapping, or use default costs
|
|
44
|
-
const normalizedModel = model.toLowerCase();
|
|
45
|
-
const modelCosts = MODEL_COSTS[normalizedModel] || MODEL_COSTS['default'];
|
|
46
|
-
// Calculate cost per token (convert from cost per 1000 tokens)
|
|
47
|
-
const promptCostPerToken = modelCosts.prompt / 1000;
|
|
48
|
-
const completionCostPerToken = modelCosts.completion / 1000;
|
|
49
|
-
return [promptCostPerToken, completionCostPerToken];
|
|
50
|
-
}
|
|
51
|
-
/**
|
|
52
|
-
* Calculate total token costs for a given model and token counts
|
|
53
|
-
*
|
|
54
|
-
* @param model Model name (e.g., 'gpt-3.5-turbo')
|
|
55
|
-
* @param promptTokens Number of prompt tokens
|
|
56
|
-
* @param completionTokens Number of completion tokens
|
|
57
|
-
* @returns Object with token counts and costs
|
|
58
|
-
*/
|
|
59
|
-
export function calculateTokenCosts(model, promptTokens, completionTokens) {
|
|
60
|
-
try {
|
|
61
|
-
// Get cost per token
|
|
62
|
-
const [promptCostPerToken, completionCostPerToken] = costPerToken(model, promptTokens, completionTokens);
|
|
63
|
-
// Calculate total costs
|
|
64
|
-
const promptCost = promptCostPerToken * promptTokens;
|
|
65
|
-
const completionCost = completionCostPerToken * completionTokens;
|
|
66
|
-
const totalCost = promptCost + completionCost;
|
|
67
|
-
return {
|
|
68
|
-
model,
|
|
69
|
-
promptTokens,
|
|
70
|
-
completionTokens,
|
|
71
|
-
totalTokens: promptTokens + completionTokens,
|
|
72
|
-
promptTokensCostUsd: promptCost,
|
|
73
|
-
completionTokensCostUsd: completionCost,
|
|
74
|
-
totalCostUsd: totalCost
|
|
75
|
-
};
|
|
76
|
-
}
|
|
77
|
-
catch (error) {
|
|
78
|
-
warn(`Error calculating token costs: ${error}`);
|
|
79
|
-
// Return default values instead of null to match Python SDK
|
|
80
|
-
return {
|
|
81
|
-
model,
|
|
82
|
-
promptTokens,
|
|
83
|
-
completionTokens,
|
|
84
|
-
totalTokens: promptTokens + completionTokens,
|
|
85
|
-
promptTokensCostUsd: 0,
|
|
86
|
-
completionTokensCostUsd: 0,
|
|
87
|
-
totalCostUsd: 0
|
|
88
|
-
};
|
|
89
|
-
}
|
|
90
|
-
}
|
|
91
|
-
//# sourceMappingURL=token-costs.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"token-costs.js","sourceRoot":"","sources":["../../../src/common/token-costs.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,IAAI,EAAE,MAAM,aAAa,CAAC;AAEnC,2CAA2C;AAC3C,6DAA6D;AAC7D,qCAAqC;AACrC,MAAM,WAAW,GAA2D;IAC1E,eAAe;IACf,OAAO,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE;IAC3C,WAAW,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE;IAC/C,aAAa,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE;IACjD,qBAAqB,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE;IACzD,sBAAsB,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE;IAC1D,QAAQ,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IAC9C,aAAa,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE;IAEpD,iBAAiB;IACjB,eAAe,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE;IACvD,mBAAmB,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IACzD,wBAAwB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE;IAE/D,gBAAgB;IAChB,kBAAkB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE;IAC1D,UAAU,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IAChD,YAAY,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IAClD,wBAAwB,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IAC9D,0BAA0B,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;IAChE,yBAAyB,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE;IAEnE,mBAAmB;IACnB,SAAS,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE;CAChD,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,UAAU,YAAY,CAC1B,KAAa,EACb,eAAuB,CAAC,EACxB,mBAA2B,CAAC;IAE5B,sDAAsD;IACtD,MAAM,eAAe,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,UAAU,GAAG,WAAW,CAAC,eAAe,CAAC,IAAI,WAAW,CAAC,SAAS,CAAC,CAAC;IAE1E,+DAA+D;IAC/D,MAAM,kBAAkB,GAAG,UAAU,CAAC,MAAM,GAAG,IAAI,CAAC;IACpD,MAAM,sBAAsB,GAAG,UAAU,CAAC,UAAU,GAAG,IAAI,CAAC;IAE5D,OAAO,CAAC,kBAAkB,EAAE,sBAAsB,CAAC,CAAC;AACtD,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CACjC,KAAa,EACb,YAAoB,EACpB,gBAAwB;IAExB,IAAI,CAAC;QACH,qBAAqB;QACrB,MAAM,CAAC,kBAAkB,EAAE,sBAAsB,CAAC,GAAG,YAAY,CAAC,KAAK,EAAE,YAAY,EAAE,gBAAgB,CAAC,CAAC;QAEzG,wBAAwB;QACxB,MAAM,UAAU,GAAG,kBAAkB,GAAG,YAAY,CAAC;QACrD,MAAM,cAAc,GAAG,sBAAsB,GAAG,gBAAgB,CAAC;QACjE,MAAM,SAAS,GAAG,UAAU,GAAG,cAAc,CAAC;QAE9C,OAAO;YACL,KAAK;YACL,YAAY;YACZ,gBAAgB;YAChB,WAAW,EAAE,YAAY,GAAG,gBAAgB;YAC5C,mBAAmB,EAAE,UAAU;YAC/B,uBAAuB,EAAE,cAAc;YACvC,YAAY,EAAE,SAAS;SACxB,CAAC;IACJ,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,CAAC,kCAAkC,KAAK,EAAE,CAAC,CAAC;QAChD,4DAA4D;QAC5D,OAAO;YACL,KAAK;YACL,YAAY;YACZ,gBAAgB;YAChB,WAAW,EAAE,YAAY,GAAG,gBAAgB;YAC5C,mBAAmB,EAAE,CAAC;YACtB,uBAAuB,EAAE,CAAC;YAC1B,YAAY,EAAE,CAAC;SAChB,CAAC;IACJ,CAAC;AACH,CAAC"}
|