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.
Files changed (408) hide show
  1. package/dist/constants.d.ts +2 -0
  2. package/dist/constants.d.ts.map +1 -0
  3. package/dist/constants.js +2 -0
  4. package/dist/constants.js.map +1 -0
  5. package/dist/data/example-params.d.ts +13 -0
  6. package/dist/data/example-params.d.ts.map +1 -0
  7. package/dist/data/example-params.js +12 -0
  8. package/dist/data/example-params.js.map +1 -0
  9. package/dist/data/example.d.ts +4 -0
  10. package/dist/data/example.d.ts.map +1 -0
  11. package/dist/data/example.js +10 -0
  12. package/dist/data/example.js.map +1 -0
  13. package/dist/data/index.d.ts +3 -0
  14. package/dist/data/index.d.ts.map +1 -0
  15. package/dist/data/index.js +3 -0
  16. package/dist/data/index.js.map +1 -0
  17. package/dist/env.d.ts +6 -0
  18. package/dist/env.d.ts.map +1 -0
  19. package/dist/env.js +13 -0
  20. package/dist/env.js.map +1 -0
  21. package/dist/index.d.ts +9 -0
  22. package/dist/index.d.ts.map +1 -0
  23. package/dist/index.js +2 -0
  24. package/dist/index.js.map +1 -0
  25. package/dist/index.mjs +2 -0
  26. package/dist/index.mjs.map +1 -0
  27. package/dist/index.umd.js +2 -0
  28. package/dist/index.umd.js.map +1 -0
  29. package/dist/internal/api/index.d.ts +23 -0
  30. package/dist/internal/api/index.d.ts.map +1 -0
  31. package/dist/internal/api/index.js +132 -0
  32. package/dist/internal/api/index.js.map +1 -0
  33. package/dist/internal/api/models/BaseScorer.d.ts +23 -0
  34. package/dist/internal/api/models/BaseScorer.d.ts.map +1 -0
  35. package/dist/internal/api/models/BaseScorer.js +6 -0
  36. package/dist/internal/api/models/BaseScorer.js.map +1 -0
  37. package/dist/internal/api/models/EvalResults.d.ts +12 -0
  38. package/dist/internal/api/models/EvalResults.d.ts.map +1 -0
  39. package/dist/internal/api/models/EvalResults.js +6 -0
  40. package/dist/internal/api/models/EvalResults.js.map +1 -0
  41. package/dist/internal/api/models/EvalResultsFetch.d.ts +9 -0
  42. package/dist/internal/api/models/EvalResultsFetch.d.ts.map +1 -0
  43. package/dist/internal/api/models/EvalResultsFetch.js +6 -0
  44. package/dist/internal/api/models/EvalResultsFetch.js.map +1 -0
  45. package/dist/internal/api/models/Example.d.ts +10 -0
  46. package/dist/internal/api/models/Example.d.ts.map +1 -0
  47. package/dist/internal/api/models/Example.js +6 -0
  48. package/dist/internal/api/models/Example.js.map +1 -0
  49. package/dist/internal/api/models/ExampleEvaluationRun.d.ts +20 -0
  50. package/dist/internal/api/models/ExampleEvaluationRun.d.ts.map +1 -0
  51. package/dist/internal/api/models/ExampleEvaluationRun.js +6 -0
  52. package/dist/internal/api/models/ExampleEvaluationRun.js.map +1 -0
  53. package/dist/internal/api/models/FetchPromptScorersRequest.d.ts +8 -0
  54. package/dist/internal/api/models/FetchPromptScorersRequest.d.ts.map +1 -0
  55. package/dist/internal/api/models/FetchPromptScorersRequest.js +6 -0
  56. package/dist/internal/api/models/FetchPromptScorersRequest.js.map +1 -0
  57. package/dist/internal/api/models/FetchPromptScorersResponse.d.ts +9 -0
  58. package/dist/internal/api/models/FetchPromptScorersResponse.d.ts.map +1 -0
  59. package/dist/internal/api/models/FetchPromptScorersResponse.js +6 -0
  60. package/dist/internal/api/models/FetchPromptScorersResponse.js.map +1 -0
  61. package/dist/internal/api/models/OtelTraceSpan.d.ts +33 -0
  62. package/dist/internal/api/models/OtelTraceSpan.d.ts.map +1 -0
  63. package/dist/internal/api/models/OtelTraceSpan.js +6 -0
  64. package/dist/internal/api/models/OtelTraceSpan.js.map +1 -0
  65. package/dist/internal/api/models/PromptScorer.d.ts +14 -0
  66. package/dist/internal/api/models/PromptScorer.d.ts.map +1 -0
  67. package/dist/internal/api/models/PromptScorer.js +6 -0
  68. package/dist/internal/api/models/PromptScorer.js.map +1 -0
  69. package/dist/internal/api/models/ResolveProjectNameRequest.d.ts +8 -0
  70. package/dist/internal/api/models/ResolveProjectNameRequest.d.ts.map +1 -0
  71. package/dist/internal/api/models/ResolveProjectNameRequest.js +6 -0
  72. package/dist/internal/api/models/ResolveProjectNameRequest.js.map +1 -0
  73. package/dist/internal/api/models/ResolveProjectNameResponse.d.ts +8 -0
  74. package/dist/internal/api/models/ResolveProjectNameResponse.d.ts.map +1 -0
  75. package/dist/internal/api/models/ResolveProjectNameResponse.js +6 -0
  76. package/dist/internal/api/models/ResolveProjectNameResponse.js.map +1 -0
  77. package/dist/internal/api/models/SavePromptScorerRequest.d.ts +12 -0
  78. package/dist/internal/api/models/SavePromptScorerRequest.d.ts.map +1 -0
  79. package/dist/internal/api/models/SavePromptScorerRequest.js +6 -0
  80. package/dist/internal/api/models/SavePromptScorerRequest.js.map +1 -0
  81. package/dist/internal/api/models/SavePromptScorerResponse.d.ts +9 -0
  82. package/dist/internal/api/models/SavePromptScorerResponse.d.ts.map +1 -0
  83. package/dist/internal/api/models/SavePromptScorerResponse.js +6 -0
  84. package/dist/internal/api/models/SavePromptScorerResponse.js.map +1 -0
  85. package/dist/internal/api/models/ScorerConfig.d.ts +13 -0
  86. package/dist/internal/api/models/ScorerConfig.d.ts.map +1 -0
  87. package/dist/internal/api/models/ScorerConfig.js +6 -0
  88. package/dist/internal/api/models/ScorerConfig.js.map +1 -0
  89. package/dist/internal/api/models/ScorerData.d.ts +17 -0
  90. package/dist/internal/api/models/ScorerData.d.ts.map +1 -0
  91. package/dist/internal/api/models/ScorerData.js +6 -0
  92. package/dist/internal/api/models/ScorerData.js.map +1 -0
  93. package/dist/internal/api/models/ScorerExistsRequest.d.ts +8 -0
  94. package/dist/internal/api/models/ScorerExistsRequest.d.ts.map +1 -0
  95. package/dist/internal/api/models/ScorerExistsRequest.js +6 -0
  96. package/dist/internal/api/models/ScorerExistsRequest.js.map +1 -0
  97. package/dist/internal/api/models/ScorerExistsResponse.d.ts +8 -0
  98. package/dist/internal/api/models/ScorerExistsResponse.d.ts.map +1 -0
  99. package/dist/internal/api/models/ScorerExistsResponse.js +6 -0
  100. package/dist/internal/api/models/ScorerExistsResponse.js.map +1 -0
  101. package/dist/internal/api/models/ScoringResult.d.ts +17 -0
  102. package/dist/internal/api/models/ScoringResult.d.ts.map +1 -0
  103. package/dist/internal/api/models/ScoringResult.js +6 -0
  104. package/dist/internal/api/models/ScoringResult.js.map +1 -0
  105. package/dist/internal/api/models/TraceEvaluationRun.d.ts +18 -0
  106. package/dist/internal/api/models/TraceEvaluationRun.d.ts.map +1 -0
  107. package/dist/internal/api/models/TraceEvaluationRun.js +6 -0
  108. package/dist/internal/api/models/TraceEvaluationRun.js.map +1 -0
  109. package/dist/internal/api/models.d.ts +24 -0
  110. package/dist/internal/api/models.d.ts.map +1 -0
  111. package/dist/internal/api/models.js +24 -0
  112. package/dist/internal/api/models.js.map +1 -0
  113. package/dist/scorers/api-scorer.d.ts +23 -0
  114. package/dist/scorers/api-scorer.d.ts.map +1 -0
  115. package/dist/scorers/api-scorer.js +36 -0
  116. package/dist/scorers/api-scorer.js.map +1 -0
  117. package/dist/scorers/api_scorers/answer-correctness-scorer.d.ts +11 -0
  118. package/dist/scorers/api_scorers/answer-correctness-scorer.d.ts.map +1 -0
  119. package/dist/scorers/api_scorers/answer-correctness-scorer.js +20 -0
  120. package/dist/scorers/api_scorers/answer-correctness-scorer.js.map +1 -0
  121. package/dist/scorers/api_scorers/answer-relevancy-scorer.d.ts +11 -0
  122. package/dist/scorers/api_scorers/answer-relevancy-scorer.d.ts.map +1 -0
  123. package/dist/scorers/api_scorers/answer-relevancy-scorer.js +19 -0
  124. package/dist/scorers/api_scorers/answer-relevancy-scorer.js.map +1 -0
  125. package/dist/scorers/api_scorers/faithfulness-scorer.d.ts +11 -0
  126. package/dist/scorers/api_scorers/faithfulness-scorer.d.ts.map +1 -0
  127. package/dist/scorers/api_scorers/faithfulness-scorer.js +20 -0
  128. package/dist/scorers/api_scorers/faithfulness-scorer.js.map +1 -0
  129. package/dist/scorers/api_scorers/index.d.ts +2 -0
  130. package/dist/scorers/api_scorers/index.d.ts.map +1 -0
  131. package/dist/scorers/api_scorers/index.js +2 -0
  132. package/dist/scorers/api_scorers/index.js.map +1 -0
  133. package/dist/scorers/api_scorers/instruction-adherence-scorer.d.ts +11 -0
  134. package/dist/scorers/api_scorers/instruction-adherence-scorer.d.ts.map +1 -0
  135. package/dist/scorers/api_scorers/instruction-adherence-scorer.js +20 -0
  136. package/dist/scorers/api_scorers/instruction-adherence-scorer.js.map +1 -0
  137. package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts +48 -0
  138. package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts.map +1 -0
  139. package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js +133 -0
  140. package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js.map +1 -0
  141. package/dist/scorers/api_scorers/prompt_scorer/index.d.ts +5 -0
  142. package/dist/scorers/api_scorers/prompt_scorer/index.d.ts.map +1 -0
  143. package/dist/scorers/api_scorers/prompt_scorer/index.js +5 -0
  144. package/dist/scorers/api_scorers/prompt_scorer/index.js.map +1 -0
  145. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.d.ts +14 -0
  146. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.d.ts.map +1 -0
  147. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js +44 -0
  148. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js.map +1 -0
  149. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.d.ts +7 -0
  150. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.d.ts.map +1 -0
  151. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js +9 -0
  152. package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js.map +1 -0
  153. package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.d.ts +6 -0
  154. package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.d.ts.map +1 -0
  155. package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js +8 -0
  156. package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js.map +1 -0
  157. package/dist/scorers/base-scorer.d.ts +10 -0
  158. package/dist/scorers/base-scorer.d.ts.map +1 -0
  159. package/dist/scorers/base-scorer.js +41 -0
  160. package/dist/scorers/base-scorer.js.map +1 -0
  161. package/dist/scorers/index.d.ts +4 -0
  162. package/dist/scorers/index.d.ts.map +1 -0
  163. package/dist/scorers/index.js +4 -0
  164. package/dist/scorers/index.js.map +1 -0
  165. package/dist/tracer/OpenTelemetryKeys.d.ts +12 -0
  166. package/dist/tracer/OpenTelemetryKeys.d.ts.map +1 -0
  167. package/dist/tracer/OpenTelemetryKeys.js +12 -0
  168. package/dist/tracer/OpenTelemetryKeys.js.map +1 -0
  169. package/dist/tracer/Tracer.d.ts +47 -0
  170. package/dist/tracer/Tracer.d.ts.map +1 -0
  171. package/dist/tracer/Tracer.js +235 -0
  172. package/dist/tracer/Tracer.js.map +1 -0
  173. package/dist/tracer/TracerConfiguration.d.ts +71 -0
  174. package/dist/tracer/TracerConfiguration.d.ts.map +1 -0
  175. package/dist/tracer/TracerConfiguration.js +111 -0
  176. package/dist/tracer/TracerConfiguration.js.map +1 -0
  177. package/dist/tracer/exporters/JudgmentSpanExporter.d.ts +36 -0
  178. package/dist/tracer/exporters/JudgmentSpanExporter.d.ts.map +1 -0
  179. package/dist/tracer/exporters/JudgmentSpanExporter.js +72 -0
  180. package/dist/tracer/exporters/JudgmentSpanExporter.js.map +1 -0
  181. package/dist/tracer/exporters/NoOpSpanExporter.d.ts +12 -0
  182. package/dist/tracer/exporters/NoOpSpanExporter.d.ts.map +1 -0
  183. package/dist/tracer/exporters/NoOpSpanExporter.js +16 -0
  184. package/dist/tracer/exporters/NoOpSpanExporter.js.map +1 -0
  185. package/dist/tracer/exporters/index.d.ts +3 -0
  186. package/dist/tracer/exporters/index.d.ts.map +1 -0
  187. package/dist/tracer/exporters/index.js +3 -0
  188. package/dist/tracer/exporters/index.js.map +1 -0
  189. package/dist/tracer/index.d.ts +5 -0
  190. package/dist/tracer/index.d.ts.map +1 -0
  191. package/dist/tracer/index.js +5 -0
  192. package/dist/tracer/index.js.map +1 -0
  193. package/dist/umd.d.ts +2 -0
  194. package/dist/umd.d.ts.map +1 -0
  195. package/dist/umd.js +2 -0
  196. package/dist/umd.js.map +1 -0
  197. package/dist/utils/annotate.d.ts +2 -0
  198. package/dist/utils/annotate.d.ts.map +1 -0
  199. package/dist/utils/annotate.js +27 -0
  200. package/dist/utils/annotate.js.map +1 -0
  201. package/dist/utils/index.d.ts +5 -0
  202. package/dist/utils/index.d.ts.map +1 -0
  203. package/dist/utils/index.js +5 -0
  204. package/dist/utils/index.js.map +1 -0
  205. package/dist/utils/logger.d.ts +27 -0
  206. package/dist/utils/logger.d.ts.map +1 -0
  207. package/dist/utils/logger.js +70 -0
  208. package/dist/utils/logger.js.map +1 -0
  209. package/dist/utils/types.d.ts +4 -0
  210. package/dist/utils/types.d.ts.map +1 -0
  211. package/dist/utils/types.js +2 -0
  212. package/dist/utils/types.js.map +1 -0
  213. package/dist/utils/validation.d.ts +11 -0
  214. package/dist/utils/validation.d.ts.map +1 -0
  215. package/dist/utils/validation.js +16 -0
  216. package/dist/utils/validation.js.map +1 -0
  217. package/dist/version.d.ts +2 -0
  218. package/dist/version.d.ts.map +1 -0
  219. package/dist/version.js +2 -0
  220. package/dist/version.js.map +1 -0
  221. package/package.json +57 -108
  222. package/LICENSE.md +0 -202
  223. package/README.md +0 -372
  224. package/dist/cjs/clients.js +0 -78
  225. package/dist/cjs/clients.js.map +0 -1
  226. package/dist/cjs/common/integrations/langgraph.js +0 -471
  227. package/dist/cjs/common/integrations/langgraph.js.map +0 -1
  228. package/dist/cjs/common/logger-instance.js +0 -64
  229. package/dist/cjs/common/logger-instance.js.map +0 -1
  230. package/dist/cjs/common/logger.js +0 -225
  231. package/dist/cjs/common/logger.js.map +0 -1
  232. package/dist/cjs/common/token-costs.js +0 -95
  233. package/dist/cjs/common/token-costs.js.map +0 -1
  234. package/dist/cjs/common/tracer.js +0 -1099
  235. package/dist/cjs/common/tracer.js.map +0 -1
  236. package/dist/cjs/constants.js +0 -347
  237. package/dist/cjs/constants.js.map +0 -1
  238. package/dist/cjs/data/datasets/eval-dataset-client.js +0 -394
  239. package/dist/cjs/data/datasets/eval-dataset-client.js.map +0 -1
  240. package/dist/cjs/data/datasets/eval-dataset.js +0 -405
  241. package/dist/cjs/data/datasets/eval-dataset.js.map +0 -1
  242. package/dist/cjs/data/example.js +0 -159
  243. package/dist/cjs/data/example.js.map +0 -1
  244. package/dist/cjs/data/result.js +0 -83
  245. package/dist/cjs/data/result.js.map +0 -1
  246. package/dist/cjs/e2etests/eval-operations.test.js +0 -282
  247. package/dist/cjs/e2etests/eval-operations.test.js.map +0 -1
  248. package/dist/cjs/e2etests/judgee-traces.test.js +0 -278
  249. package/dist/cjs/e2etests/judgee-traces.test.js.map +0 -1
  250. package/dist/cjs/evaluation-run.js +0 -136
  251. package/dist/cjs/evaluation-run.js.map +0 -1
  252. package/dist/cjs/exporters/otel-exporter.js +0 -352
  253. package/dist/cjs/exporters/otel-exporter.js.map +0 -1
  254. package/dist/cjs/index.js +0 -78
  255. package/dist/cjs/index.js.map +0 -1
  256. package/dist/cjs/judges/index.js +0 -237
  257. package/dist/cjs/judges/index.js.map +0 -1
  258. package/dist/cjs/judgment-client.js +0 -798
  259. package/dist/cjs/judgment-client.js.map +0 -1
  260. package/dist/cjs/rules.js +0 -322
  261. package/dist/cjs/rules.js.map +0 -1
  262. package/dist/cjs/run-evaluation.js +0 -700
  263. package/dist/cjs/run-evaluation.js.map +0 -1
  264. package/dist/cjs/scorers/api-scorer.js +0 -304
  265. package/dist/cjs/scorers/api-scorer.js.map +0 -1
  266. package/dist/cjs/scorers/base-scorer.js +0 -293
  267. package/dist/cjs/scorers/base-scorer.js.map +0 -1
  268. package/dist/cjs/scorers/metrics/answer-correctness/answer-correctness.js +0 -562
  269. package/dist/cjs/scorers/metrics/answer-correctness/answer-correctness.js.map +0 -1
  270. package/dist/cjs/scorers/metrics/answer-correctness/index.js +0 -19
  271. package/dist/cjs/scorers/metrics/answer-correctness/index.js.map +0 -1
  272. package/dist/cjs/scorers/metrics/answer-correctness/prompts.js +0 -175
  273. package/dist/cjs/scorers/metrics/answer-correctness/prompts.js.map +0 -1
  274. package/dist/cjs/scorers/metrics/answer-relevancy/answer-relevancy.js +0 -514
  275. package/dist/cjs/scorers/metrics/answer-relevancy/answer-relevancy.js.map +0 -1
  276. package/dist/cjs/scorers/metrics/answer-relevancy/index.js +0 -19
  277. package/dist/cjs/scorers/metrics/answer-relevancy/index.js.map +0 -1
  278. package/dist/cjs/scorers/metrics/answer-relevancy/prompts.js +0 -179
  279. package/dist/cjs/scorers/metrics/answer-relevancy/prompts.js.map +0 -1
  280. package/dist/cjs/scorers/metrics/faithfulness/faithfulness.js +0 -512
  281. package/dist/cjs/scorers/metrics/faithfulness/faithfulness.js.map +0 -1
  282. package/dist/cjs/scorers/metrics/faithfulness/index.js +0 -19
  283. package/dist/cjs/scorers/metrics/faithfulness/index.js.map +0 -1
  284. package/dist/cjs/scorers/metrics/faithfulness/prompts.js +0 -232
  285. package/dist/cjs/scorers/metrics/faithfulness/prompts.js.map +0 -1
  286. package/dist/cjs/scorers/metrics/hallucination/hallucination.js +0 -386
  287. package/dist/cjs/scorers/metrics/hallucination/hallucination.js.map +0 -1
  288. package/dist/cjs/scorers/metrics/hallucination/index.js +0 -11
  289. package/dist/cjs/scorers/metrics/hallucination/index.js.map +0 -1
  290. package/dist/cjs/scorers/metrics/hallucination/prompts.js +0 -106
  291. package/dist/cjs/scorers/metrics/hallucination/prompts.js.map +0 -1
  292. package/dist/cjs/scorers/metrics/instruction-adherence/index.js +0 -19
  293. package/dist/cjs/scorers/metrics/instruction-adherence/index.js.map +0 -1
  294. package/dist/cjs/scorers/metrics/instruction-adherence/instruction-adherence.js +0 -378
  295. package/dist/cjs/scorers/metrics/instruction-adherence/instruction-adherence.js.map +0 -1
  296. package/dist/cjs/scorers/metrics/instruction-adherence/prompts.js +0 -124
  297. package/dist/cjs/scorers/metrics/instruction-adherence/prompts.js.map +0 -1
  298. package/dist/esm/clients.js +0 -47
  299. package/dist/esm/clients.js.map +0 -1
  300. package/dist/esm/common/integrations/langgraph.js +0 -444
  301. package/dist/esm/common/integrations/langgraph.js.map +0 -1
  302. package/dist/esm/common/logger-instance.js +0 -59
  303. package/dist/esm/common/logger-instance.js.map +0 -1
  304. package/dist/esm/common/logger.js +0 -208
  305. package/dist/esm/common/logger.js.map +0 -1
  306. package/dist/esm/common/token-costs.js +0 -91
  307. package/dist/esm/common/token-costs.js.map +0 -1
  308. package/dist/esm/common/tracer.js +0 -1093
  309. package/dist/esm/common/tracer.js.map +0 -1
  310. package/dist/esm/constants.js +0 -341
  311. package/dist/esm/constants.js.map +0 -1
  312. package/dist/esm/data/datasets/eval-dataset-client.js +0 -387
  313. package/dist/esm/data/datasets/eval-dataset-client.js.map +0 -1
  314. package/dist/esm/data/datasets/eval-dataset.js +0 -375
  315. package/dist/esm/data/datasets/eval-dataset.js.map +0 -1
  316. package/dist/esm/data/example.js +0 -154
  317. package/dist/esm/data/example.js.map +0 -1
  318. package/dist/esm/data/result.js +0 -78
  319. package/dist/esm/data/result.js.map +0 -1
  320. package/dist/esm/e2etests/eval-operations.test.js +0 -254
  321. package/dist/esm/e2etests/eval-operations.test.js.map +0 -1
  322. package/dist/esm/e2etests/judgee-traces.test.js +0 -253
  323. package/dist/esm/e2etests/judgee-traces.test.js.map +0 -1
  324. package/dist/esm/evaluation-run.js +0 -132
  325. package/dist/esm/evaluation-run.js.map +0 -1
  326. package/dist/esm/exporters/otel-exporter.js +0 -348
  327. package/dist/esm/exporters/otel-exporter.js.map +0 -1
  328. package/dist/esm/index.js +0 -20
  329. package/dist/esm/index.js.map +0 -1
  330. package/dist/esm/judges/index.js +0 -205
  331. package/dist/esm/judges/index.js.map +0 -1
  332. package/dist/esm/judgment-client.js +0 -768
  333. package/dist/esm/judgment-client.js.map +0 -1
  334. package/dist/esm/rules.js +0 -314
  335. package/dist/esm/rules.js.map +0 -1
  336. package/dist/esm/run-evaluation.js +0 -681
  337. package/dist/esm/run-evaluation.js.map +0 -1
  338. package/dist/esm/scorers/api-scorer.js +0 -286
  339. package/dist/esm/scorers/api-scorer.js.map +0 -1
  340. package/dist/esm/scorers/base-scorer.js +0 -287
  341. package/dist/esm/scorers/base-scorer.js.map +0 -1
  342. package/dist/esm/scorers/metrics/answer-correctness/answer-correctness.js +0 -558
  343. package/dist/esm/scorers/metrics/answer-correctness/answer-correctness.js.map +0 -1
  344. package/dist/esm/scorers/metrics/answer-correctness/index.js +0 -3
  345. package/dist/esm/scorers/metrics/answer-correctness/index.js.map +0 -1
  346. package/dist/esm/scorers/metrics/answer-correctness/prompts.js +0 -171
  347. package/dist/esm/scorers/metrics/answer-correctness/prompts.js.map +0 -1
  348. package/dist/esm/scorers/metrics/answer-relevancy/answer-relevancy.js +0 -510
  349. package/dist/esm/scorers/metrics/answer-relevancy/answer-relevancy.js.map +0 -1
  350. package/dist/esm/scorers/metrics/answer-relevancy/index.js +0 -3
  351. package/dist/esm/scorers/metrics/answer-relevancy/index.js.map +0 -1
  352. package/dist/esm/scorers/metrics/answer-relevancy/prompts.js +0 -175
  353. package/dist/esm/scorers/metrics/answer-relevancy/prompts.js.map +0 -1
  354. package/dist/esm/scorers/metrics/faithfulness/faithfulness.js +0 -508
  355. package/dist/esm/scorers/metrics/faithfulness/faithfulness.js.map +0 -1
  356. package/dist/esm/scorers/metrics/faithfulness/index.js +0 -3
  357. package/dist/esm/scorers/metrics/faithfulness/index.js.map +0 -1
  358. package/dist/esm/scorers/metrics/faithfulness/prompts.js +0 -228
  359. package/dist/esm/scorers/metrics/faithfulness/prompts.js.map +0 -1
  360. package/dist/esm/scorers/metrics/hallucination/hallucination.js +0 -382
  361. package/dist/esm/scorers/metrics/hallucination/hallucination.js.map +0 -1
  362. package/dist/esm/scorers/metrics/hallucination/index.js +0 -3
  363. package/dist/esm/scorers/metrics/hallucination/index.js.map +0 -1
  364. package/dist/esm/scorers/metrics/hallucination/prompts.js +0 -102
  365. package/dist/esm/scorers/metrics/hallucination/prompts.js.map +0 -1
  366. package/dist/esm/scorers/metrics/instruction-adherence/index.js +0 -3
  367. package/dist/esm/scorers/metrics/instruction-adherence/index.js.map +0 -1
  368. package/dist/esm/scorers/metrics/instruction-adherence/instruction-adherence.js +0 -374
  369. package/dist/esm/scorers/metrics/instruction-adherence/instruction-adherence.js.map +0 -1
  370. package/dist/esm/scorers/metrics/instruction-adherence/prompts.js +0 -120
  371. package/dist/esm/scorers/metrics/instruction-adherence/prompts.js.map +0 -1
  372. package/dist/types/clients.d.ts +0 -6
  373. package/dist/types/common/integrations/langgraph.d.ts +0 -39
  374. package/dist/types/common/logger-instance.d.ts +0 -3
  375. package/dist/types/common/logger.d.ts +0 -54
  376. package/dist/types/common/token-costs.d.ts +0 -36
  377. package/dist/types/common/tracer.d.ts +0 -249
  378. package/dist/types/constants.d.ts +0 -51
  379. package/dist/types/data/datasets/eval-dataset-client.d.ts +0 -44
  380. package/dist/types/data/datasets/eval-dataset.d.ts +0 -45
  381. package/dist/types/data/example.d.ts +0 -82
  382. package/dist/types/data/result.d.ts +0 -51
  383. package/dist/types/e2etests/eval-operations.test.d.ts +0 -5
  384. package/dist/types/e2etests/judgee-traces.test.d.ts +0 -5
  385. package/dist/types/evaluation-run.d.ts +0 -44
  386. package/dist/types/exporters/otel-exporter.d.ts +0 -16
  387. package/dist/types/index.d.ts +0 -11
  388. package/dist/types/judges/index.d.ts +0 -50
  389. package/dist/types/judgment-client.d.ts +0 -149
  390. package/dist/types/rules.d.ts +0 -120
  391. package/dist/types/run-evaluation.d.ts +0 -78
  392. package/dist/types/scorers/api-scorer.d.ts +0 -84
  393. package/dist/types/scorers/base-scorer.d.ts +0 -162
  394. package/dist/types/scorers/metrics/answer-correctness/answer-correctness.d.ts +0 -80
  395. package/dist/types/scorers/metrics/answer-correctness/index.d.ts +0 -2
  396. package/dist/types/scorers/metrics/answer-correctness/prompts.d.ts +0 -71
  397. package/dist/types/scorers/metrics/answer-relevancy/answer-relevancy.d.ts +0 -74
  398. package/dist/types/scorers/metrics/answer-relevancy/index.d.ts +0 -2
  399. package/dist/types/scorers/metrics/answer-relevancy/prompts.d.ts +0 -71
  400. package/dist/types/scorers/metrics/faithfulness/faithfulness.d.ts +0 -73
  401. package/dist/types/scorers/metrics/faithfulness/index.d.ts +0 -2
  402. package/dist/types/scorers/metrics/faithfulness/prompts.d.ts +0 -94
  403. package/dist/types/scorers/metrics/hallucination/hallucination.d.ts +0 -63
  404. package/dist/types/scorers/metrics/hallucination/index.d.ts +0 -3
  405. package/dist/types/scorers/metrics/hallucination/prompts.d.ts +0 -63
  406. package/dist/types/scorers/metrics/instruction-adherence/index.d.ts +0 -2
  407. package/dist/types/scorers/metrics/instruction-adherence/instruction-adherence.d.ts +0 -63
  408. package/dist/types/scorers/metrics/instruction-adherence/prompts.d.ts +0 -78
@@ -1,558 +0,0 @@
1
- var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
2
- function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
3
- return new (P || (P = Promise))(function (resolve, reject) {
4
- function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
5
- function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
6
- function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
7
- step((generator = generator.apply(thisArg, _arguments || [])).next());
8
- });
9
- };
10
- import { log, info, warn, error } from '../../../common/logger.js';
11
- import { JudgevalScorer } from '../../base-scorer.js';
12
- import { APIScorer } from '../../../constants.js';
13
- import { AnswerCorrectnessTemplate, StatementsSchema, VerdictsSchema, ReasonSchema } from './prompts.js';
14
- import { createJudge } from '../../../judges/index.js';
15
- // Required parameters for this scorer
16
- const requiredParams = ['input', 'actualOutput', 'expectedOutput'];
17
- /**
18
- * Answer Correctness Scorer
19
- *
20
- * This scorer evaluates whether the actual output correctly represents
21
- * the expected output by extracting statements from the expected output
22
- * and checking if they are correctly represented in the actual output.
23
- */
24
- export class AnswerCorrectnessScorer extends JudgevalScorer {
25
- /**
26
- * Constructor for AnswerCorrectnessScorer
27
- * @param threshold Minimum score to consider the evaluation successful (default: 0.5)
28
- * @param model LLM to use for evaluation (string or Judge instance)
29
- * @param include_reason Whether to generate a reason for the score
30
- * @param async_mode Whether to use asynchronous evaluation
31
- * @param strict_mode If true, sets threshold to 1.0 (requiring perfect match)
32
- * @param verbose_mode Enables detailed logging
33
- * @param user Optional user identifier for the LLM
34
- * @param additional_metadata Additional metadata to include in the result
35
- */
36
- constructor(threshold = 0.5, model, include_reason = true, async_mode = true, strict_mode = false, verbose_mode = true, user, additional_metadata) {
37
- super(APIScorer.ANSWER_CORRECTNESS, strict_mode ? 1.0 : threshold, additional_metadata, include_reason, async_mode, strict_mode, verbose_mode);
38
- info(`Initializing AnswerCorrectnessScorer with threshold=${this.threshold}, model=${model}, strict_mode=${strict_mode}`);
39
- const { judge, usingNativeModel } = createJudge(model, user);
40
- this.model = judge;
41
- this.usingNativeModel = usingNativeModel;
42
- this.evaluation_model = this.model.getModelName();
43
- log(`Using model: ${this.evaluation_model}`);
44
- // Set required fields for this scorer
45
- this.requiredFields = ['input', 'actualOutput', 'expectedOutput'];
46
- }
47
- /**
48
- * Get statements from expected output asynchronously
49
- */
50
- _aGetStatements(expectedOutput) {
51
- return __awaiter(this, void 0, void 0, function* () {
52
- log("Getting statements asynchronously");
53
- // Handle string array
54
- const expectedOutputStr = Array.isArray(expectedOutput) ? expectedOutput.join('\n') : expectedOutput;
55
- const prompt = AnswerCorrectnessTemplate.deduceStatements(expectedOutputStr);
56
- try {
57
- const response = yield this.model.aGenerate(prompt);
58
- // Parse the response
59
- try {
60
- const jsonResponse = JSON.parse(response);
61
- const parsed = StatementsSchema.safeParse(jsonResponse);
62
- if (parsed.success) {
63
- return parsed.data.statements;
64
- }
65
- else {
66
- // Fallback to direct access if schema validation fails
67
- warn("Schema validation failed, falling back to raw response parsing");
68
- if (jsonResponse.statements && Array.isArray(jsonResponse.statements)) {
69
- return jsonResponse.statements;
70
- }
71
- }
72
- }
73
- catch (parseError) {
74
- warn(`Error parsing JSON response: ${parseError}`);
75
- // Try to extract JSON from the response text
76
- const jsonMatch = response.match(/\{[\s\S]*\}/);
77
- if (jsonMatch) {
78
- try {
79
- const extractedJson = JSON.parse(jsonMatch[0]);
80
- if (extractedJson.statements && Array.isArray(extractedJson.statements)) {
81
- return extractedJson.statements;
82
- }
83
- }
84
- catch (e) {
85
- error(`Failed to extract JSON from response: ${e}`);
86
- }
87
- }
88
- }
89
- // If all parsing attempts fail, return empty array
90
- error("Failed to parse statements from model response");
91
- return [];
92
- }
93
- catch (e) {
94
- error(`Error getting statements: ${e}`);
95
- return [];
96
- }
97
- });
98
- }
99
- /**
100
- * Get statements from expected output synchronously
101
- */
102
- _getStatements(expectedOutput) {
103
- // Handle string array
104
- const expectedOutputStr = Array.isArray(expectedOutput) ? expectedOutput.join('\n') : expectedOutput;
105
- const prompt = AnswerCorrectnessTemplate.deduceStatements(expectedOutputStr);
106
- try {
107
- const response = this.model.generate(prompt);
108
- // Parse the response
109
- try {
110
- const jsonResponse = JSON.parse(response);
111
- const parsed = StatementsSchema.safeParse(jsonResponse);
112
- if (parsed.success) {
113
- return parsed.data.statements;
114
- }
115
- else {
116
- // Fallback to direct access if schema validation fails
117
- warn("Schema validation failed, falling back to raw response parsing");
118
- if (jsonResponse.statements && Array.isArray(jsonResponse.statements)) {
119
- return jsonResponse.statements;
120
- }
121
- }
122
- }
123
- catch (parseError) {
124
- warn(`Error parsing JSON response: ${parseError}`);
125
- // Try to extract JSON from the response text
126
- const jsonMatch = response.match(/\{[\s\S]*\}/);
127
- if (jsonMatch) {
128
- try {
129
- const extractedJson = JSON.parse(jsonMatch[0]);
130
- if (extractedJson.statements && Array.isArray(extractedJson.statements)) {
131
- return extractedJson.statements;
132
- }
133
- }
134
- catch (e) {
135
- error(`Failed to extract JSON from response: ${e}`);
136
- }
137
- }
138
- }
139
- // If all parsing attempts fail, return empty array
140
- error("Failed to parse statements from model response");
141
- return [];
142
- }
143
- catch (e) {
144
- error(`Error getting statements: ${e}`);
145
- return [];
146
- }
147
- }
148
- /**
149
- * Get verdicts for statements against actual output asynchronously
150
- */
151
- _aGetVerdicts(actualOutput) {
152
- return __awaiter(this, void 0, void 0, function* () {
153
- log("Getting verdicts asynchronously");
154
- if (!this.statements || this.statements.length === 0) {
155
- warn("No statements to evaluate");
156
- return [];
157
- }
158
- // Handle string array
159
- const actualOutputStr = Array.isArray(actualOutput) ? actualOutput.join('\n') : actualOutput;
160
- const prompt = AnswerCorrectnessTemplate.generateVerdicts(this.statements, actualOutputStr);
161
- try {
162
- const response = yield this.model.aGenerate(prompt);
163
- // Parse the response
164
- try {
165
- const jsonResponse = JSON.parse(response);
166
- const parsed = VerdictsSchema.safeParse(jsonResponse);
167
- if (parsed.success) {
168
- return parsed.data.verdicts;
169
- }
170
- else {
171
- // Fallback to direct access if schema validation fails
172
- warn("Schema validation failed, falling back to raw response parsing");
173
- if (jsonResponse.verdicts && Array.isArray(jsonResponse.verdicts)) {
174
- return jsonResponse.verdicts.map((v) => ({
175
- verdict: v.verdict,
176
- reason: v.reason
177
- }));
178
- }
179
- }
180
- }
181
- catch (parseError) {
182
- warn(`Error parsing JSON response: ${parseError}`);
183
- // Try to extract JSON from the response text
184
- const jsonMatch = response.match(/\{[\s\S]*\}/);
185
- if (jsonMatch) {
186
- try {
187
- const extractedJson = JSON.parse(jsonMatch[0]);
188
- if (extractedJson.verdicts && Array.isArray(extractedJson.verdicts)) {
189
- return extractedJson.verdicts.map((v) => ({
190
- verdict: v.verdict,
191
- reason: v.reason
192
- }));
193
- }
194
- }
195
- catch (e) {
196
- error(`Failed to extract JSON from response: ${e}`);
197
- }
198
- }
199
- }
200
- // If all parsing attempts fail, return empty array
201
- error("Failed to parse verdicts from model response");
202
- return [];
203
- }
204
- catch (e) {
205
- error(`Error getting verdicts: ${e}`);
206
- return [];
207
- }
208
- });
209
- }
210
- /**
211
- * Get verdicts for statements against actual output synchronously
212
- */
213
- _getVerdicts(actualOutput) {
214
- if (!this.statements || this.statements.length === 0) {
215
- warn("No statements to evaluate");
216
- return [];
217
- }
218
- // Handle string array
219
- const actualOutputStr = Array.isArray(actualOutput) ? actualOutput.join('\n') : actualOutput;
220
- const prompt = AnswerCorrectnessTemplate.generateVerdicts(this.statements, actualOutputStr);
221
- try {
222
- const response = this.model.generate(prompt);
223
- // Parse the response
224
- try {
225
- const jsonResponse = JSON.parse(response);
226
- const parsed = VerdictsSchema.safeParse(jsonResponse);
227
- if (parsed.success) {
228
- return parsed.data.verdicts;
229
- }
230
- else {
231
- // Fallback to direct access if schema validation fails
232
- warn("Schema validation failed, falling back to raw response parsing");
233
- if (jsonResponse.verdicts && Array.isArray(jsonResponse.verdicts)) {
234
- return jsonResponse.verdicts.map((v) => ({
235
- verdict: v.verdict,
236
- reason: v.reason
237
- }));
238
- }
239
- }
240
- }
241
- catch (parseError) {
242
- warn(`Error parsing JSON response: ${parseError}`);
243
- // Try to extract JSON from the response text
244
- const jsonMatch = response.match(/\{[\s\S]*\}/);
245
- if (jsonMatch) {
246
- try {
247
- const extractedJson = JSON.parse(jsonMatch[0]);
248
- if (extractedJson.verdicts && Array.isArray(extractedJson.verdicts)) {
249
- return extractedJson.verdicts.map((v) => ({
250
- verdict: v.verdict,
251
- reason: v.reason
252
- }));
253
- }
254
- }
255
- catch (e) {
256
- error(`Failed to extract JSON from response: ${e}`);
257
- }
258
- }
259
- }
260
- // If all parsing attempts fail, return empty array
261
- error("Failed to parse verdicts from model response");
262
- return [];
263
- }
264
- catch (e) {
265
- error(`Error getting verdicts: ${e}`);
266
- return [];
267
- }
268
- }
269
- /**
270
- * Get reason for the score asynchronously
271
- */
272
- _aGetReason() {
273
- return __awaiter(this, void 0, void 0, function* () {
274
- var _a;
275
- if (!this.include_reason) {
276
- return undefined;
277
- }
278
- if (!this.verdicts || this.verdicts.length === 0) {
279
- return undefined;
280
- }
281
- try {
282
- // Get incorrect statements with their verdicts
283
- const incorrectStatements = [];
284
- for (let i = 0; i < this.statements.length; i++) {
285
- if (i < this.verdicts.length && this.verdicts[i].verdict.toLowerCase() === "no") {
286
- incorrectStatements.push([this.statements[i], this.verdicts[i].reason]);
287
- }
288
- }
289
- if (incorrectStatements.length === 0) {
290
- return "All statements in the expected output are correctly represented in the actual output.";
291
- }
292
- // Generate reason
293
- const prompt = AnswerCorrectnessTemplate.generateReason(incorrectStatements, ((_a = this.score) === null || _a === void 0 ? void 0 : _a.toString()) || "0");
294
- const reasonText = yield this.model.aGenerate(prompt);
295
- const parsedReason = ReasonSchema.safeParse(JSON.parse(reasonText));
296
- if (!parsedReason.success) {
297
- error(`Failed to parse reason: ${parsedReason.error}`);
298
- return undefined;
299
- }
300
- return parsedReason.data.reason;
301
- }
302
- catch (err) {
303
- error(`Error getting reason: ${err}`);
304
- return undefined;
305
- }
306
- });
307
- }
308
- /**
309
- * Get reason for the score synchronously
310
- */
311
- _getReason() {
312
- var _a;
313
- if (!this.include_reason) {
314
- return undefined;
315
- }
316
- if (!this.verdicts || this.verdicts.length === 0) {
317
- return undefined;
318
- }
319
- try {
320
- // Get incorrect statements with their verdicts
321
- const incorrectStatements = [];
322
- for (let i = 0; i < this.statements.length; i++) {
323
- if (i < this.verdicts.length && this.verdicts[i].verdict.toLowerCase() === "no") {
324
- incorrectStatements.push([this.statements[i], this.verdicts[i].reason]);
325
- }
326
- }
327
- if (incorrectStatements.length === 0) {
328
- return "All statements in the expected output are correctly represented in the actual output.";
329
- }
330
- // Generate reason
331
- const prompt = AnswerCorrectnessTemplate.generateReason(incorrectStatements, ((_a = this.score) === null || _a === void 0 ? void 0 : _a.toString()) || "0");
332
- const reasonText = this.model.generate(prompt);
333
- const parsedReason = ReasonSchema.safeParse(JSON.parse(reasonText));
334
- if (!parsedReason.success) {
335
- error(`Failed to parse reason: ${parsedReason.error}`);
336
- return undefined;
337
- }
338
- return parsedReason.data.reason;
339
- }
340
- catch (err) {
341
- error(`Error getting reason: ${err}`);
342
- return undefined;
343
- }
344
- }
345
- /**
346
- * Compute score based on verdicts
347
- */
348
- _computeScore() {
349
- log("Computing score");
350
- // If we have no statements or verdicts due to API errors, return 0 instead of 1
351
- // This ensures that when API calls fail, we don't incorrectly return a perfect score
352
- if (!this.statements || this.statements.length === 0) {
353
- return 0;
354
- }
355
- if (!this.verdicts || this.verdicts.length === 0) {
356
- return 0;
357
- }
358
- let correctCount = 0;
359
- for (const verdict of this.verdicts) {
360
- if (verdict.verdict.trim().toLowerCase() === "yes") {
361
- correctCount++;
362
- }
363
- }
364
- const score = correctCount / this.verdicts.length;
365
- // Match Python implementation's handling of strict_mode
366
- return this.strict_mode && score < this.threshold ? 0 : score;
367
- }
368
- /**
369
- * Check if example has required parameters
370
- */
371
- _checkExampleParams(example) {
372
- for (const param of this.requiredFields) {
373
- if (param === 'input' && !example.input) {
374
- throw new Error(`Example is missing required parameter: input`);
375
- }
376
- else if (param === 'actualOutput' && !example.actualOutput) {
377
- throw new Error(`Example is missing required parameter: actualOutput`);
378
- }
379
- else if (param === 'expectedOutput' && !example.expectedOutput) {
380
- throw new Error(`Example is missing required parameter: expectedOutput`);
381
- }
382
- else if (param === 'context' && (!example.context || !Array.isArray(example.context))) {
383
- throw new Error(`Example is missing required parameter: context (must be an array)`);
384
- }
385
- else if (param === 'retrievalContext' && (!example.retrievalContext || !Array.isArray(example.retrievalContext))) {
386
- throw new Error(`Example is missing required parameter: retrievalContext (must be an array)`);
387
- }
388
- }
389
- }
390
- /**
391
- * Create verbose logs for debugging
392
- */
393
- _createVerboseLogs() {
394
- if (!this.verbose_mode) {
395
- return '';
396
- }
397
- const steps = [
398
- `Statements:\n${JSON.stringify(this.statements, null, 2)}`,
399
- `Verdicts:\n${JSON.stringify(this.verdicts, null, 2)}`,
400
- `Score: ${this.score}\nReason: ${this.reason}`
401
- ];
402
- return steps.join('\n\n');
403
- }
404
- /**
405
- * Score an example synchronously - this is for compatibility with the Python SDK
406
- */
407
- syncScoreExample(example) {
408
- info("Starting example scoring (sync mode)");
409
- try {
410
- // Check required parameters
411
- this._checkExampleParams(example);
412
- // Process example
413
- if (this.async_mode) {
414
- throw new Error("Cannot use synchronous scoreExample with async_mode=true. Use async scoreExample instead.");
415
- }
416
- // Track token usage
417
- let promptTokens = 0;
418
- let completionTokens = 0;
419
- // Get statements and track tokens
420
- this.statements = this._getStatements(example.expectedOutput);
421
- promptTokens += 500; // Approximate tokens for statements prompt
422
- completionTokens += 100; // Approximate tokens for statements response
423
- // Get verdicts and track tokens
424
- this.verdicts = this._getVerdicts(example.actualOutput);
425
- promptTokens += 800; // Approximate tokens for verdicts prompt
426
- completionTokens += 200; // Approximate tokens for verdicts response
427
- // Compute score
428
- this.score = this._computeScore();
429
- // Get reason if needed
430
- if (this.include_reason) {
431
- this.reason = this._getReason();
432
- promptTokens += 300; // Approximate tokens for reason prompt
433
- completionTokens += 100; // Approximate tokens for reason response
434
- }
435
- // Calculate evaluation cost
436
- this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
437
- // Set success flag
438
- this.success = this.score >= this.threshold;
439
- // Create verbose logs if needed
440
- if (this.verbose_mode) {
441
- this.verbose_logs = this._createVerboseLogs();
442
- }
443
- info(`Scoring completed with score: ${this.score}`);
444
- // Ensure all fields match the ScorerData interface
445
- return {
446
- name: this.type,
447
- threshold: this.threshold,
448
- success: this.success || false,
449
- score: this.score || 0,
450
- reason: this.reason !== undefined ? this.reason : null,
451
- strict_mode: this.strict_mode || false,
452
- evaluation_model: this.evaluation_model || null,
453
- error: null,
454
- evaluation_cost: this.evaluation_cost,
455
- verbose_logs: this.verbose_logs ? this.verbose_logs : null,
456
- additional_metadata: this.additional_metadata || {}
457
- };
458
- }
459
- catch (error) {
460
- // Handle errors
461
- const errorMessage = error instanceof Error ? error.message : String(error);
462
- this.error = errorMessage;
463
- this.success = false;
464
- return {
465
- name: this.type,
466
- threshold: this.threshold,
467
- success: false,
468
- score: 0,
469
- reason: `Error during scoring: ${errorMessage}`,
470
- strict_mode: this.strict_mode || false,
471
- evaluation_model: this.evaluation_model || null,
472
- error: errorMessage,
473
- evaluation_cost: null,
474
- verbose_logs: null,
475
- additional_metadata: this.additional_metadata || {}
476
- };
477
- }
478
- }
479
- /**
480
- * Score an example - this is the main method that should be called
481
- * It will use async or sync methods based on the async_mode setting
482
- */
483
- scoreExample(example) {
484
- return __awaiter(this, void 0, void 0, function* () {
485
- if (!this.async_mode) {
486
- return this.syncScoreExample(example);
487
- }
488
- info("Starting example scoring (async mode)");
489
- try {
490
- // Check required parameters
491
- this._checkExampleParams(example);
492
- // Track token usage
493
- let promptTokens = 0;
494
- let completionTokens = 0;
495
- // Process example
496
- this.statements = yield this._aGetStatements(example.expectedOutput);
497
- promptTokens += 500; // Approximate tokens for statements prompt
498
- completionTokens += 100; // Approximate tokens for statements response
499
- this.verdicts = yield this._aGetVerdicts(example.actualOutput);
500
- promptTokens += 800; // Approximate tokens for verdicts prompt
501
- completionTokens += 200; // Approximate tokens for verdicts response
502
- this.score = this._computeScore();
503
- if (this.include_reason) {
504
- this.reason = yield this._aGetReason();
505
- promptTokens += 300; // Approximate tokens for reason prompt
506
- completionTokens += 100; // Approximate tokens for reason response
507
- }
508
- this.success = this._successCheck();
509
- if (this.verbose_mode) {
510
- this.verbose_logs = this._createVerboseLogs();
511
- }
512
- // Calculate evaluation cost
513
- this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
514
- info(`Scoring completed with score: ${this.score}`);
515
- // Ensure all fields match the ScorerData interface
516
- return {
517
- name: this.type,
518
- threshold: this.threshold,
519
- success: this.success || false,
520
- score: this.score || 0,
521
- reason: this.reason !== undefined ? this.reason : null,
522
- strict_mode: this.strict_mode || false,
523
- evaluation_model: this.evaluation_model || null,
524
- error: null,
525
- evaluation_cost: this.evaluation_cost,
526
- verbose_logs: this.verbose_logs ? this.verbose_logs : null,
527
- additional_metadata: this.additional_metadata || {}
528
- };
529
- }
530
- catch (error) {
531
- // Handle errors
532
- const errorMessage = error instanceof Error ? error.message : String(error);
533
- this.error = errorMessage;
534
- this.success = false;
535
- return {
536
- name: this.type,
537
- threshold: this.threshold,
538
- success: false,
539
- score: 0,
540
- reason: `Error during scoring: ${errorMessage}`,
541
- strict_mode: this.strict_mode || false,
542
- evaluation_model: this.evaluation_model || null,
543
- error: errorMessage,
544
- evaluation_cost: null,
545
- verbose_logs: null,
546
- additional_metadata: this.additional_metadata || {}
547
- };
548
- }
549
- });
550
- }
551
- /**
552
- * Get the name of the scorer
553
- */
554
- get name() {
555
- return "Answer Correctness";
556
- }
557
- }
558
- //# 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,OAAO,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,2BAA2B,CAAC;AACnE,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAGtD,OAAO,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAClD,OAAO,EAAE,yBAAyB,EAAa,gBAAgB,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AACpH,OAAO,EAAS,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAG9D,sCAAsC;AACtC,MAAM,cAAc,GAAG,CAAC,OAAO,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;AAEnE;;;;;;GAMG;AACH,MAAM,OAAO,uBAAwB,SAAQ,cAAc;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,SAAS,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,IAAI,CAAC,uDAAuD,IAAI,CAAC,SAAS,WAAW,KAAK,iBAAiB,WAAW,EAAE,CAAC,CAAC;QAE1H,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,GAAG,WAAW,CAAC,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,GAAG,CAAC,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,GAAG,CAAC,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,yBAAyB,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,gBAAgB,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,IAAI,CAAC,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,IAAI,CAAC,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,KAAK,CAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;wBACtD,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,mDAAmD;gBACnD,KAAK,CAAC,gDAAgD,CAAC,CAAC;gBACxD,OAAO,EAAE,CAAC;YACZ,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,KAAK,CAAC,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,yBAAyB,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,gBAAgB,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,IAAI,CAAC,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,IAAI,CAAC,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,KAAK,CAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;oBACtD,CAAC;gBACH,CAAC;YACH,CAAC;YAED,mDAAmD;YACnD,KAAK,CAAC,gDAAgD,CAAC,CAAC;YACxD,OAAO,EAAE,CAAC;QACZ,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,KAAK,CAAC,6BAA6B,CAAC,EAAE,CAAC,CAAC;YACxC,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC;IAED;;OAEG;IACW,aAAa,CAAC,YAA+B;;YACzD,GAAG,CAAC,iCAAiC,CAAC,CAAC;YAEvC,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACrD,IAAI,CAAC,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,yBAAyB,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,cAAc,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,IAAI,CAAC,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,IAAI,CAAC,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,KAAK,CAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;wBACtD,CAAC;oBACH,CAAC;gBACH,CAAC;gBAED,mDAAmD;gBACnD,KAAK,CAAC,8CAA8C,CAAC,CAAC;gBACtD,OAAO,EAAE,CAAC;YACZ,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,KAAK,CAAC,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,IAAI,CAAC,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,yBAAyB,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,cAAc,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,IAAI,CAAC,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,IAAI,CAAC,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,KAAK,CAAC,yCAAyC,CAAC,EAAE,CAAC,CAAC;oBACtD,CAAC;gBACH,CAAC;YACH,CAAC;YAED,mDAAmD;YACnD,KAAK,CAAC,8CAA8C,CAAC,CAAC;YACtD,OAAO,EAAE,CAAC;QACZ,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,KAAK,CAAC,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,yBAAyB,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,YAAY,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;gBAEpE,IAAI,CAAC,YAAY,CAAC,OAAO,EAAE,CAAC;oBAC1B,KAAK,CAAC,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,KAAK,CAAC,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,yBAAyB,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,YAAY,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;YAEpE,IAAI,CAAC,YAAY,CAAC,OAAO,EAAE,CAAC;gBAC1B,KAAK,CAAC,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,KAAK,CAAC,yBAAyB,GAAG,EAAE,CAAC,CAAC;YACtC,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAED;;OAEG;IACK,aAAa;QACnB,GAAG,CAAC,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,IAAI,CAAC,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,IAAI,CAAC,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,IAAI,CAAC,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,IAAI,CAAC,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"}
@@ -1,3 +0,0 @@
1
- export * from './answer-correctness.js';
2
- export * from './prompts.js';
3
- //# 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,cAAc,yBAAyB,CAAC;AACxC,cAAc,cAAc,CAAC"}