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
@@ -1,228 +0,0 @@
1
- /**
2
- * Utility prompts for FaithfulnessScorer
3
- */
4
- import { z } from 'zod';
5
- // Schema definitions for LLM responses
6
- export const FaithfulnessVerdictSchema = z.object({
7
- verdict: z.string(),
8
- reason: z.string().optional()
9
- });
10
- export const VerdictsSchema = z.object({
11
- verdicts: z.array(FaithfulnessVerdictSchema)
12
- });
13
- export const TruthsSchema = z.object({
14
- truths: z.array(z.string())
15
- });
16
- export const ClaimsSchema = z.object({
17
- claims: z.array(z.object({
18
- claim: z.string(),
19
- quote: z.string()
20
- }))
21
- });
22
- export const ReasonSchema = z.object({
23
- reason: z.string()
24
- });
25
- /**
26
- * Template prompts for the FaithfulnessScorer
27
- */
28
- export class FaithfulnessTemplate {
29
- /**
30
- * Generate a prompt to extract claims from the actual output
31
- */
32
- static findClaims(text, allClaims = false) {
33
- return `==== TASK INSTRUCTIONS ====
34
- You will be provided with a passage of text. Based on the text, your task is to generate a comprehensive list of ALL CLAIMS that can be inferred from the text.
35
- For every claim that you derive from the text, provide the source of the claim via quoting the original text. Please try to extract EVERY CLAIM that is in the original text; priortize generating the most claims rather than being concise.
36
- You should NOT include any prior knowledge, and take the text at face value when extracting claims.
37
-
38
- ==== FORMATTING YOUR ANSWER ====
39
- Please return your answer in JSON format, with the "claims" key as a list of JSON objects with keys "claim" and "quote". No words or explanation beyond the output JSON is needed.
40
-
41
- ==== EXAMPLES ====
42
-
43
- ---- START OF EXAMPLE 1 ----
44
- Example Text:
45
- "Einstein won the nobel prize in 1968 for his discovery of the photoelectric effect."
46
-
47
- Example JSON:
48
- {
49
- "claims": [
50
- {
51
- "claim": "Einstein won the nobel prize for his discovery of the photoelectric effect.",
52
- "quote": "Einstein won the nobel prize in 1968 for his discovery of the photoelectric effect."
53
- },
54
- {
55
- "claim": "Einstein won the nobel prize in 1968.",
56
- "quote": "Einstein won the nobel prize in 1968 for his discovery of the photoelectric effect."
57
- }
58
- ]
59
- }
60
- ---- END OF EXAMPLE 1 ----
61
-
62
- ---- START OF EXAMPLE 2 ----
63
- Example Text: "The Wright brothers successfully flew the first powered airplane on December 17, 1903, in Kitty Hawk, North Carolina."
64
-
65
- {
66
- "claims": [
67
- {
68
- "claim": "The Wright brothers flew the first powered airplane.",
69
- "quote": "The Wright brothers successfully flew the first powered airplane on December 17, 1903, in Kitty Hawk, North Carolina."
70
- },
71
- {
72
- "claim": "The Wright brothers made their flight in Kitty Hawk, North Carolina.",
73
- "quote": "The Wright brothers successfully flew the first powered airplane on December 17, 1903, in Kitty Hawk, North Carolina."
74
- },
75
- {
76
- "claim": "The first powered airplane flight occurred on December 17, 1903.",
77
- "quote": "The Wright brothers successfully flew the first powered airplane on December 17, 1903, in Kitty Hawk, North Carolina."
78
- }
79
- ]
80
- }
81
- ---- END OF EXAMPLE 2 ----
82
-
83
- ---- START OF EXAMPLE 3 ----
84
- Example Text:
85
- "The Great Wall of China was built over many centuries by different Chinese dynasties. Construction began more than 2,000 years ago during the Warring States period. The most famous sections were built during the Ming Dynasty. The wall stretches for thousands of miles across northern China and was primarily built for military defense."
86
-
87
- Example JSON:
88
- {
89
- "claims": [
90
- {
91
- "claim": "The Great Wall of China was built by multiple Chinese dynasties",
92
- "quote": "The Great Wall of China was built over many centuries by different Chinese dynasties."
93
- },
94
- {
95
- "claim": "Construction of the Great Wall began over 2,000 years ago",
96
- "quote": "Construction began more than 2,000 years ago during the Warring States period."
97
- },
98
- {
99
- "claim": "Construction started during the Warring States period",
100
- "quote": "Construction began more than 2,000 years ago during the Warring States period."
101
- },
102
- {
103
- "claim": "The most well-known parts of the wall were constructed during the Ming Dynasty",
104
- "quote": "The most famous sections were built during the Ming Dynasty."
105
- },
106
- {
107
- "claim": "The Great Wall extends for thousands of miles across northern China",
108
- "quote": "The wall stretches for thousands of miles across northern China and was primarily built for military defense."
109
- },
110
- {
111
- "claim": "The Great Wall was mainly constructed for defense purposes",
112
- "quote": "The wall stretches for thousands of miles across northern China and was primarily built for military defense."
113
- }
114
- ]
115
- }
116
- ---- END OF EXAMPLE 3 ----
117
-
118
- ==== YOUR TURN ====
119
- Text:
120
- ${text}
121
-
122
- JSON:`;
123
- }
124
- /**
125
- * Generate a prompt to evaluate claims against the retrieval context
126
- */
127
- static generateVerdicts(claims, retrievalContext) {
128
- const claimsStr = claims.map((claim, i) => `Claim ${i + 1}: ${claim}`).join('\n');
129
- return `==== TASK INSTRUCTIONS ====
130
- You will be provided with a set of claims and a retrieval context. Your task is to determine whether each claim is supported by the retrieval context.
131
-
132
- For each claim, provide a verdict of "yes" if the claim is supported by the retrieval context, "no" if the claim is contradicted by the retrieval context, or "partially" if the claim is partially supported.
133
-
134
- Also provide a reason for your verdict, citing evidence from the retrieval context.
135
-
136
- ==== FORMATTING YOUR ANSWER ====
137
- Please return your answer in JSON format, with the "verdicts" key as a list of JSON objects with keys "verdict" and "reason". The verdict should be one of "yes", "no", or "partially". The reason should explain your verdict.
138
-
139
- ==== EXAMPLES ====
140
-
141
- ---- START OF EXAMPLE 1 ----
142
- Example Claims:
143
- Claim 1: Einstein won the Nobel Prize in 1921.
144
- Claim 2: Einstein won the Nobel Prize for his work on relativity.
145
-
146
- Example Retrieval Context:
147
- "Albert Einstein was awarded the Nobel Prize in Physics in 1921 for his explanation of the photoelectric effect, not for his theories of relativity which were still controversial at the time."
148
-
149
- Example JSON:
150
- {
151
- "verdicts": [
152
- {
153
- "verdict": "yes",
154
- "reason": "The retrieval context explicitly states that 'Albert Einstein was awarded the Nobel Prize in Physics in 1921'."
155
- },
156
- {
157
- "verdict": "no",
158
- "reason": "The retrieval context contradicts this claim, stating that Einstein won the Nobel Prize 'for his explanation of the photoelectric effect, not for his theories of relativity'."
159
- }
160
- ]
161
- }
162
- ---- END OF EXAMPLE 1 ----
163
-
164
- ---- START OF EXAMPLE 2 ----
165
- Example Claims:
166
- Claim 1: The Great Wall of China is visible from space.
167
- Claim 2: The Great Wall of China was built during the Ming Dynasty.
168
-
169
- Example Retrieval Context:
170
- "The Great Wall of China was built over many centuries by different Chinese dynasties, with the most famous sections built during the Ming Dynasty (1368-1644). Contrary to popular belief, the Great Wall is not visible from space with the naked eye, as confirmed by multiple astronauts."
171
-
172
- Example JSON:
173
- {
174
- "verdicts": [
175
- {
176
- "verdict": "no",
177
- "reason": "The retrieval context explicitly contradicts this claim, stating that 'the Great Wall is not visible from space with the naked eye, as confirmed by multiple astronauts'."
178
- },
179
- {
180
- "verdict": "partially",
181
- "reason": "The retrieval context states that 'the most famous sections [were] built during the Ming Dynasty', but also mentions that the wall 'was built over many centuries by different Chinese dynasties'. This suggests that while significant portions were built during the Ming Dynasty, the entire wall was not built during this period."
182
- }
183
- ]
184
- }
185
- ---- END OF EXAMPLE 2 ----
186
-
187
- ==== YOUR TURN ====
188
- Claims:
189
- ${claimsStr}
190
-
191
- Retrieval Context:
192
- ${retrievalContext}
193
-
194
- JSON:`;
195
- }
196
- /**
197
- * Generate a prompt to explain the score based on verdicts
198
- */
199
- static generateReason(verdicts, score) {
200
- const verdictsStr = verdicts.map((v, i) => {
201
- return `Verdict ${i + 1}: ${v.verdict}\nReason: ${v.reason || 'No reason provided'}\n------`;
202
- }).join('\n');
203
- return `==== TASK INSTRUCTIONS ====
204
- You will be provided with a faithfulness score and a list of verdicts for claims made in a model's output. Your task is to provide a CLEAR and CONCISE reason for the faithfulness score.
205
-
206
- The faithfulness score represents how well the model's output is supported by the provided retrieval context. A score of 1.0 means all claims are fully supported, while a score of 0.0 means none of the claims are supported.
207
-
208
- You should explain why the score is what it is, highlighting key supported and unsupported claims.
209
-
210
- ==== FORMATTING YOUR ANSWER ====
211
- IMPORTANT: Please make sure to only return in JSON format, with the 'reason' key providing the reason.
212
- Example JSON:
213
- {
214
- "reason": "The faithfulness score is <score> because <your_reason>."
215
- }
216
-
217
- ==== YOUR TURN ====
218
- ---- FAITHFULNESS SCORE ----
219
- ${score}
220
-
221
- ---- VERDICTS ----
222
- ${verdictsStr}
223
-
224
- ---- YOUR RESPONSE ----
225
- JSON:`;
226
- }
227
- }
228
- //# sourceMappingURL=prompts.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"prompts.js","sourceRoot":"","sources":["../../../../../src/scorers/metrics/faithfulness/prompts.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,uCAAuC;AACvC,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC9B,CAAC,CAAC;AAIH,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IACrC,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,yBAAyB,CAAC;CAC7C,CAAC,CAAC;AAIH,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IACnC,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;CAC5B,CAAC,CAAC;AAIH,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IACnC,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;QACvB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;QACjB,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;KAClB,CAAC,CAAC;CACJ,CAAC,CAAC;AAIH,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IACnC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC;AAIH;;GAEG;AACH,MAAM,OAAO,oBAAoB;IAC/B;;OAEG;IACH,MAAM,CAAC,UAAU,CAAC,IAAY,EAAE,YAAqB,KAAK;QACxD,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAuFT,IAAI;;MAEA,CAAC;IACL,CAAC;IAED;;OAEG;IACH,MAAM,CAAC,gBAAgB,CAAC,MAAgB,EAAE,gBAAwB;QAChE,MAAM,SAAS,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,GAAC,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEhF,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA4DT,SAAS;;;EAGT,gBAAgB;;MAEZ,CAAC;IACL,CAAC;IAED;;OAEG;IACH,MAAM,CAAC,cAAc,CAAC,QAA+B,EAAE,KAAa;QAClE,MAAM,WAAW,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACxC,OAAO,WAAW,CAAC,GAAC,CAAC,KAAK,CAAC,CAAC,OAAO,aAAa,CAAC,CAAC,MAAM,IAAI,oBAAoB,UAAU,CAAC;QAC7F,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,OAAO;;;;;;;;;;;;;;;;EAgBT,KAAK;;;EAGL,WAAW;;;MAGP,CAAC;IACL,CAAC;CACF"}
@@ -1,382 +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 { JudgevalScorer } from '../../base-scorer.js';
11
- import { APIScorer } from '../../../constants.js';
12
- import { info } from '../../../common/logger.js';
13
- import { HallucinationTemplate, VerdictsSchema, ReasonSchema } from './prompts.js';
14
- import { createJudge } from '../../../judges/index.js';
15
- // Required parameters for this scorer
16
- const required_params = ['actualOutput', 'context'];
17
- /**
18
- * HallucinationScorer evaluates whether an LLM's output contains hallucinations
19
- * by comparing it against provided context.
20
- *
21
- * The score is the fraction of context segments that contradict the output.
22
- * Lower scores are better (0 = no hallucinations, 1 = all contexts contradict the output).
23
- */
24
- export class HallucinationScorer extends JudgevalScorer {
25
- /**
26
- * Create a new HallucinationScorer
27
- *
28
- * @param threshold - Success threshold (default: 0.5)
29
- * @param model - Model to use for evaluation (default: DefaultJudge)
30
- * @param include_reason - Whether to include a reason for the score (default: true)
31
- * @param async_mode - Whether to use async mode (default: false)
32
- * @param strict_mode - Whether to use strict mode (default: false)
33
- * @param verbose_mode - Whether to include verbose logs (default: false)
34
- */
35
- constructor(threshold = 0.5, model = undefined, include_reason = true, async_mode = false, strict_mode = false, verbose_mode = false) {
36
- super(APIScorer.HALLUCINATION, strict_mode ? 1 : threshold, undefined, include_reason, async_mode, strict_mode, verbose_mode);
37
- this._verdicts = [];
38
- const { judge, usingNativeModel } = createJudge(model);
39
- this.model = judge;
40
- this.using_native_model = usingNativeModel;
41
- this.evaluation_model = this.model.getModelName();
42
- this.requiredFields = ['actualOutput', 'context'];
43
- }
44
- /**
45
- * Generate verdicts for each context
46
- */
47
- _aGenerateVerdicts(actualOutput, contexts) {
48
- return __awaiter(this, void 0, void 0, function* () {
49
- const prompt = HallucinationTemplate.generateVerdicts(actualOutput, contexts);
50
- if (this.using_native_model) {
51
- const res = yield this.model.aGenerate(prompt);
52
- try {
53
- const data = JSON.parse(res);
54
- return data.verdicts.map((item) => ({
55
- verdict: item.verdict,
56
- reason: item.reason
57
- }));
58
- }
59
- catch (error) {
60
- throw new Error(`Failed to parse response: ${error}`);
61
- }
62
- }
63
- else {
64
- try {
65
- const parseVerdictsResponse = (response) => {
66
- const parsed = JSON.parse(response);
67
- const result = VerdictsSchema.safeParse(parsed);
68
- if (result.success) {
69
- return result.data;
70
- }
71
- throw new Error(`Invalid response format: ${result.error}`);
72
- };
73
- const res = yield this.model.aGenerate(prompt);
74
- return parseVerdictsResponse(res).verdicts;
75
- }
76
- catch (error) {
77
- const res = yield this.model.aGenerate(prompt);
78
- try {
79
- const data = JSON.parse(res);
80
- return data.verdicts.map((item) => ({
81
- verdict: item.verdict,
82
- reason: item.reason
83
- }));
84
- }
85
- catch (parseError) {
86
- throw new Error(`Failed to parse response: ${parseError}`);
87
- }
88
- }
89
- }
90
- });
91
- }
92
- /**
93
- * Generate verdicts for each context (synchronous)
94
- */
95
- _generateVerdicts(actualOutput, contexts) {
96
- const prompt = HallucinationTemplate.generateVerdicts(actualOutput, contexts);
97
- if (this.using_native_model) {
98
- const res = this.model.generate(prompt);
99
- try {
100
- const data = JSON.parse(res);
101
- return data.verdicts.map((item) => ({
102
- verdict: item.verdict,
103
- reason: item.reason
104
- }));
105
- }
106
- catch (error) {
107
- throw new Error(`Failed to parse response: ${error}`);
108
- }
109
- }
110
- else {
111
- try {
112
- const parseVerdictsResponse = (response) => {
113
- const parsed = JSON.parse(response);
114
- const result = VerdictsSchema.safeParse(parsed);
115
- if (result.success) {
116
- return result.data;
117
- }
118
- throw new Error(`Invalid response format: ${result.error}`);
119
- };
120
- const res = this.model.generate(prompt);
121
- return parseVerdictsResponse(res).verdicts;
122
- }
123
- catch (error) {
124
- const res = this.model.generate(prompt);
125
- try {
126
- const data = JSON.parse(res);
127
- return data.verdicts.map((item) => ({
128
- verdict: item.verdict,
129
- reason: item.reason
130
- }));
131
- }
132
- catch (parseError) {
133
- throw new Error(`Failed to parse response: ${parseError}`);
134
- }
135
- }
136
- }
137
- }
138
- /**
139
- * Generate a reason for the score
140
- */
141
- _aGenerateReason(actualOutput, contexts) {
142
- return __awaiter(this, void 0, void 0, function* () {
143
- if (!this.include_reason) {
144
- return "No reason provided (include_reason is false)";
145
- }
146
- const prompt = HallucinationTemplate.generateReason(actualOutput, contexts);
147
- if (this.using_native_model) {
148
- const res = yield this.model.aGenerate(prompt);
149
- try {
150
- const data = JSON.parse(res);
151
- return data.reason || "No reason provided in response";
152
- }
153
- catch (error) {
154
- throw new Error(`Failed to parse response: ${error}`);
155
- }
156
- }
157
- else {
158
- try {
159
- const parseReasonResponse = (response) => {
160
- const parsed = JSON.parse(response);
161
- const result = ReasonSchema.safeParse(parsed);
162
- if (result.success) {
163
- return result.data;
164
- }
165
- throw new Error(`Invalid response format: ${result.error}`);
166
- };
167
- const res = yield this.model.aGenerate(prompt);
168
- return parseReasonResponse(res).reason;
169
- }
170
- catch (error) {
171
- const res = yield this.model.aGenerate(prompt);
172
- try {
173
- const data = JSON.parse(res);
174
- return data.reason || "No reason provided in response";
175
- }
176
- catch (parseError) {
177
- throw new Error(`Failed to parse response: ${parseError}`);
178
- }
179
- }
180
- }
181
- });
182
- }
183
- /**
184
- * Generate a reason for the score (synchronous)
185
- */
186
- _generateReason(actualOutput, contexts) {
187
- if (!this.include_reason) {
188
- return "No reason provided (include_reason is false)";
189
- }
190
- const prompt = HallucinationTemplate.generateReason(actualOutput, contexts);
191
- if (this.using_native_model) {
192
- const res = this.model.generate(prompt);
193
- try {
194
- const data = JSON.parse(res);
195
- return data.reason || "No reason provided in response";
196
- }
197
- catch (error) {
198
- throw new Error(`Failed to parse response: ${error}`);
199
- }
200
- }
201
- else {
202
- try {
203
- const parseReasonResponse = (response) => {
204
- const parsed = JSON.parse(response);
205
- const result = ReasonSchema.safeParse(parsed);
206
- if (result.success) {
207
- return result.data;
208
- }
209
- throw new Error(`Invalid response format: ${result.error}`);
210
- };
211
- const res = this.model.generate(prompt);
212
- return parseReasonResponse(res).reason;
213
- }
214
- catch (error) {
215
- const res = this.model.generate(prompt);
216
- try {
217
- const data = JSON.parse(res);
218
- return data.reason || "No reason provided in response";
219
- }
220
- catch (parseError) {
221
- throw new Error(`Failed to parse response: ${parseError}`);
222
- }
223
- }
224
- }
225
- }
226
- /**
227
- * Calculate the hallucination score
228
- */
229
- _computeScore() {
230
- if (this._verdicts.length === 0) {
231
- return 0;
232
- }
233
- let contradictions = 0;
234
- for (const verdict of this._verdicts) {
235
- if (verdict.verdict.trim().toLowerCase() === "no") {
236
- contradictions += 1;
237
- }
238
- }
239
- return contradictions / this._verdicts.length;
240
- }
241
- /**
242
- * Create verbose logs for debugging
243
- */
244
- _createVerboseLogs() {
245
- if (!this.verbose_mode) {
246
- return null;
247
- }
248
- const steps = [
249
- `Verdicts:\n${JSON.stringify(this._verdicts, null, 2)}`,
250
- `Score: ${this.score}\nReason: ${this.reason || "No reason provided"}`
251
- ];
252
- return steps.join('\n\n');
253
- }
254
- /**
255
- * Score an example synchronously
256
- */
257
- syncScoreExample(example) {
258
- info("Starting example scoring (sync mode)");
259
- try {
260
- // Check required parameters
261
- this._checkExampleParams(example);
262
- // Process example
263
- const contexts = Array.isArray(example.context) ? example.context : [example.context || ''];
264
- this._verdicts = this._generateVerdicts(example.actualOutput, contexts);
265
- // Calculate score
266
- this.score = this._computeScore();
267
- this.reason = this._generateReason(example.actualOutput, contexts) || '';
268
- this.success = this._successCheck();
269
- // Create verbose logs if enabled
270
- const verbose_logs = this._createVerboseLogs();
271
- // Calculate evaluation cost
272
- const promptTokens = 600; // Estimate - in a real implementation, track actual tokens
273
- const completionTokens = 200; // Estimate - in a real implementation, track actual tokens
274
- this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
275
- info(`Scoring completed with score: ${this.score}`);
276
- // Return ScorerData object
277
- return {
278
- name: this.type,
279
- threshold: this.threshold,
280
- success: this.success,
281
- score: this.score,
282
- reason: this.reason,
283
- strict_mode: this.strict_mode,
284
- evaluation_model: this.evaluation_model || null,
285
- error: null,
286
- evaluation_cost: this.evaluation_cost || null,
287
- verbose_logs: verbose_logs,
288
- additional_metadata: {
289
- verdicts: this._verdicts
290
- }
291
- };
292
- }
293
- catch (error) {
294
- // Handle errors
295
- const errorMessage = error instanceof Error ? error.message : String(error);
296
- this.error = errorMessage;
297
- return {
298
- name: this.type,
299
- threshold: this.threshold,
300
- success: false,
301
- score: 0,
302
- reason: `Error during scoring: ${errorMessage}`,
303
- strict_mode: this.strict_mode,
304
- evaluation_model: this.evaluation_model || null,
305
- error: errorMessage,
306
- evaluation_cost: null,
307
- verbose_logs: null,
308
- additional_metadata: {}
309
- };
310
- }
311
- }
312
- /**
313
- * Score an example asynchronously
314
- */
315
- scoreExample(example) {
316
- return __awaiter(this, void 0, void 0, function* () {
317
- if (!this.async_mode) {
318
- return this.syncScoreExample(example);
319
- }
320
- info("Starting example scoring (async mode)");
321
- try {
322
- // Check required parameters
323
- this._checkExampleParams(example);
324
- // Process example
325
- const contexts = Array.isArray(example.context) ? example.context : [example.context || ''];
326
- this._verdicts = yield this._aGenerateVerdicts(example.actualOutput, contexts);
327
- // Calculate score
328
- this.score = this._computeScore();
329
- this.reason = (yield this._aGenerateReason(example.actualOutput, contexts)) || '';
330
- this.success = this._successCheck();
331
- // Create verbose logs if enabled
332
- const verbose_logs = this._createVerboseLogs();
333
- // Calculate evaluation cost
334
- const promptTokens = 600; // Estimate - in a real implementation, track actual tokens
335
- const completionTokens = 200; // Estimate - in a real implementation, track actual tokens
336
- this.evaluation_cost = this._calculateTokenCosts(this.evaluation_model || 'gpt-3.5-turbo', promptTokens, completionTokens);
337
- info(`Scoring completed with score: ${this.score}`);
338
- // Return ScorerData object
339
- return {
340
- name: this.type,
341
- threshold: this.threshold,
342
- success: this.success,
343
- score: this.score,
344
- reason: this.reason,
345
- strict_mode: this.strict_mode,
346
- evaluation_model: this.evaluation_model || null,
347
- error: null,
348
- evaluation_cost: this.evaluation_cost || null,
349
- verbose_logs: verbose_logs,
350
- additional_metadata: {
351
- verdicts: this._verdicts
352
- }
353
- };
354
- }
355
- catch (error) {
356
- // Handle errors
357
- const errorMessage = error instanceof Error ? error.message : String(error);
358
- this.error = errorMessage;
359
- return {
360
- name: this.type,
361
- threshold: this.threshold,
362
- success: false,
363
- score: 0,
364
- reason: `Error during scoring: ${errorMessage}`,
365
- strict_mode: this.strict_mode,
366
- evaluation_model: this.evaluation_model || null,
367
- error: errorMessage,
368
- evaluation_cost: null,
369
- verbose_logs: null,
370
- additional_metadata: {}
371
- };
372
- }
373
- });
374
- }
375
- /**
376
- * Get the name of the scorer
377
- */
378
- get name() {
379
- return "Hallucination";
380
- }
381
- }
382
- //# sourceMappingURL=hallucination.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"hallucination.js","sourceRoot":"","sources":["../../../../../src/scorers/metrics/hallucination/hallucination.ts"],"names":[],"mappings":";;;;;;;;;AAEA,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAClD,OAAO,EAAO,IAAI,EAAe,MAAM,2BAA2B,CAAC;AACnE,OAAO,EACL,qBAAqB,EAErB,cAAc,EACd,YAAY,EACb,MAAM,cAAc,CAAC;AACtB,OAAO,EAAS,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAE9D,sCAAsC;AACtC,MAAM,eAAe,GAAG,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;AAEpD;;;;;;GAMG;AACH,MAAM,OAAO,mBAAoB,SAAQ,cAAc;IAKrD;;;;;;;;;OASG;IACH,YACE,YAAoB,GAAG,EACvB,QAAoC,SAAS,EAC7C,iBAA0B,IAAI,EAC9B,aAAsB,KAAK,EAC3B,cAAuB,KAAK,EAC5B,eAAwB,KAAK;QAE7B,KAAK,CACH,SAAS,CAAC,aAAa,EACvB,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,EAC3B,SAAS,EACT,cAAc,EACd,UAAU,EACV,WAAW,EACX,YAAY,CACb,CAAC;QA5BI,cAAS,GAA2B,EAAE,CAAC;QA8B7C,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC;QACvD,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,kBAAkB,GAAG,gBAAgB,CAAC;QAC3C,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC;QAClD,IAAI,CAAC,cAAc,GAAG,CAAC,cAAc,EAAE,SAAS,CAAC,CAAC;IACpD,CAAC;IAED;;OAEG;IACW,kBAAkB,CAAC,YAAoB,EAAE,QAAkB;;YACvE,MAAM,MAAM,GAAG,qBAAqB,CAAC,gBAAgB,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;YAE9E,IAAI,IAAI,CAAC,kBAAkB,EAAE,CAAC;gBAC5B,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAC/C,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;oBAC7B,OAAO,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE,CAAC,CAAC;wBACvC,OAAO,EAAE,IAAI,CAAC,OAAO;wBACrB,MAAM,EAAE,IAAI,CAAC,MAAM;qBACpB,CAAC,CAAC,CAAC;gBACN,CAAC;gBAAC,OAAO,KAAK,EAAE,CAAC;oBACf,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;gBACxD,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,IAAI,CAAC;oBACH,MAAM,qBAAqB,GAAG,CAAC,QAAgB,EAAwC,EAAE;wBACvF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;wBACpC,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;wBAChD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;4BACnB,OAAO,MAAM,CAAC,IAAI,CAAC;wBACrB,CAAC;wBACD,MAAM,IAAI,KAAK,CAAC,4BAA4B,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;oBAC9D,CAAC,CAAC;oBAEF,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAC/C,OAAO,qBAAqB,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC;gBAC7C,CAAC;gBAAC,OAAO,KAAK,EAAE,CAAC;oBACf,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAC/C,IAAI,CAAC;wBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;wBAC7B,OAAO,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE,CAAC,CAAC;4BACvC,OAAO,EAAE,IAAI,CAAC,OAAO;4BACrB,MAAM,EAAE,IAAI,CAAC,MAAM;yBACpB,CAAC,CAAC,CAAC;oBACN,CAAC;oBAAC,OAAO,UAAU,EAAE,CAAC;wBACpB,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,EAAE,CAAC,CAAC;oBAC7D,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACK,iBAAiB,CAAC,YAAoB,EAAE,QAAkB;QAChE,MAAM,MAAM,GAAG,qBAAqB,CAAC,gBAAgB,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;QAE9E,IAAI,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAC5B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YACxC,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;gBAC7B,OAAO,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE,CAAC,CAAC;oBACvC,OAAO,EAAE,IAAI,CAAC,OAAO;oBACrB,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB,CAAC,CAAC,CAAC;YACN,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;YACxD,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC;gBACH,MAAM,qBAAqB,GAAG,CAAC,QAAgB,EAAwC,EAAE;oBACvF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;oBACpC,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAChD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;wBACnB,OAAO,MAAM,CAAC,IAAI,CAAC;oBACrB,CAAC;oBACD,MAAM,IAAI,KAAK,CAAC,4BAA4B,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;gBAC9D,CAAC,CAAC;gBAEF,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;gBACxC,OAAO,qBAAqB,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAC;YAC7C,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;gBACxC,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;oBAC7B,OAAO,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAS,EAAE,EAAE,CAAC,CAAC;wBACvC,OAAO,EAAE,IAAI,CAAC,OAAO;wBACrB,MAAM,EAAE,IAAI,CAAC,MAAM;qBACpB,CAAC,CAAC,CAAC;gBACN,CAAC;gBAAC,OAAO,UAAU,EAAE,CAAC;oBACpB,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,EAAE,CAAC,CAAC;gBAC7D,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACW,gBAAgB,CAAC,YAAoB,EAAE,QAAkB;;YACrE,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;gBACzB,OAAO,8CAA8C,CAAC;YACxD,CAAC;YAED,MAAM,MAAM,GAAG,qBAAqB,CAAC,cAAc,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;YAE5E,IAAI,IAAI,CAAC,kBAAkB,EAAE,CAAC;gBAC5B,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;gBAC/C,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;oBAC7B,OAAO,IAAI,CAAC,MAAM,IAAI,gCAAgC,CAAC;gBACzD,CAAC;gBAAC,OAAO,KAAK,EAAE,CAAC;oBACf,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;gBACxD,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,IAAI,CAAC;oBACH,MAAM,mBAAmB,GAAG,CAAC,QAAgB,EAAsB,EAAE;wBACnE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;wBACpC,MAAM,MAAM,GAAG,YAAY,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;wBAC9C,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;4BACnB,OAAO,MAAM,CAAC,IAAI,CAAC;wBACrB,CAAC;wBACD,MAAM,IAAI,KAAK,CAAC,4BAA4B,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;oBAC9D,CAAC,CAAC;oBAEF,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAC/C,OAAO,mBAAmB,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC;gBACzC,CAAC;gBAAC,OAAO,KAAK,EAAE,CAAC;oBACf,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAC/C,IAAI,CAAC;wBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;wBAC7B,OAAO,IAAI,CAAC,MAAM,IAAI,gCAAgC,CAAC;oBACzD,CAAC;oBAAC,OAAO,UAAU,EAAE,CAAC;wBACpB,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,EAAE,CAAC,CAAC;oBAC7D,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACK,eAAe,CAAC,YAAoB,EAAE,QAAkB;QAC9D,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;YACzB,OAAO,8CAA8C,CAAC;QACxD,CAAC;QAED,MAAM,MAAM,GAAG,qBAAqB,CAAC,cAAc,CAAC,YAAY,EAAE,QAAQ,CAAC,CAAC;QAE5E,IAAI,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAC5B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;YACxC,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;gBAC7B,OAAO,IAAI,CAAC,MAAM,IAAI,gCAAgC,CAAC;YACzD,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,IAAI,KAAK,CAAC,6BAA6B,KAAK,EAAE,CAAC,CAAC;YACxD,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC;gBACH,MAAM,mBAAmB,GAAG,CAAC,QAAgB,EAAsB,EAAE;oBACnE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;oBACpC,MAAM,MAAM,GAAG,YAAY,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;oBAC9C,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;wBACnB,OAAO,MAAM,CAAC,IAAI,CAAC;oBACrB,CAAC;oBACD,MAAM,IAAI,KAAK,CAAC,4BAA4B,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;gBAC9D,CAAC,CAAC;gBAEF,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;gBACxC,OAAO,mBAAmB,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC;YACzC,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;gBACxC,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;oBAC7B,OAAO,IAAI,CAAC,MAAM,IAAI,gCAAgC,CAAC;gBACzD,CAAC;gBAAC,OAAO,UAAU,EAAE,CAAC;oBACpB,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,EAAE,CAAC,CAAC;gBAC7D,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED;;OAEG;IACK,aAAa;QACnB,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAChC,OAAO,CAAC,CAAC;QACX,CAAC;QAED,IAAI,cAAc,GAAG,CAAC,CAAC;QACvB,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACrC,IAAI,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,KAAK,IAAI,EAAE,CAAC;gBAClD,cAAc,IAAI,CAAC,CAAC;YACtB,CAAC;QACH,CAAC;QAED,OAAO,cAAc,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;IAChD,CAAC;IAED;;OAEG;IACK,kBAAkB;QACxB,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,OAAO,IAAI,CAAC;QACd,CAAC;QAED,MAAM,KAAK,GAAG;YACZ,cAAc,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE;YACvD,UAAU,IAAI,CAAC,KAAK,aAAa,IAAI,CAAC,MAAM,IAAI,oBAAoB,EAAE;SACvE,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,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;YAC5F,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,iBAAiB,CAAC,OAAO,CAAC,YAAsB,EAAE,QAAQ,CAAC,CAAC;YAElF,kBAAkB;YAClB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;YAClC,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,eAAe,CAAC,OAAO,CAAC,YAAsB,EAAE,QAAQ,CAAC,IAAI,EAAE,CAAC;YACnF,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;YAEpC,iCAAiC;YACjC,MAAM,YAAY,GAAG,IAAI,CAAC,kBAAkB,EAAE,CAAC;YAE/C,4BAA4B;YAC5B,MAAM,YAAY,GAAG,GAAG,CAAC,CAAC,2DAA2D;YACrF,MAAM,gBAAgB,GAAG,GAAG,CAAC,CAAC,2DAA2D;YACzF,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,oBAAoB,CAC9C,IAAI,CAAC,gBAAgB,IAAI,eAAe,EACxC,YAAY,EACZ,gBAAgB,CACjB,CAAC;YAEF,IAAI,CAAC,iCAAiC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;YAEpD,2BAA2B;YAC3B,OAAO;gBACL,IAAI,EAAE,IAAI,CAAC,IAAI;gBACf,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,MAAM,EAAE,IAAI,CAAC,MAAM;gBACnB,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAC/C,KAAK,EAAE,IAAI;gBACX,eAAe,EAAE,IAAI,CAAC,eAAe,IAAI,IAAI;gBAC7C,YAAY,EAAE,YAAY;gBAC1B,mBAAmB,EAAE;oBACnB,QAAQ,EAAE,IAAI,CAAC,SAAS;iBACzB;aACF,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;YAE1B,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;gBAC7B,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAC/C,KAAK,EAAE,YAAY;gBACnB,eAAe,EAAE,IAAI;gBACrB,YAAY,EAAE,IAAI;gBAClB,mBAAmB,EAAE,EAAE;aACxB,CAAC;QACJ,CAAC;IACH,CAAC;IAED;;OAEG;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,kBAAkB;gBAClB,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAC5F,IAAI,CAAC,SAAS,GAAG,MAAM,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,YAAsB,EAAE,QAAQ,CAAC,CAAC;gBAEzF,kBAAkB;gBAClB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;gBAClC,IAAI,CAAC,MAAM,GAAG,CAAA,MAAM,IAAI,CAAC,gBAAgB,CAAC,OAAO,CAAC,YAAsB,EAAE,QAAQ,CAAC,KAAI,EAAE,CAAC;gBAC1F,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;gBAEpC,iCAAiC;gBACjC,MAAM,YAAY,GAAG,IAAI,CAAC,kBAAkB,EAAE,CAAC;gBAE/C,4BAA4B;gBAC5B,MAAM,YAAY,GAAG,GAAG,CAAC,CAAC,2DAA2D;gBACrF,MAAM,gBAAgB,GAAG,GAAG,CAAC,CAAC,2DAA2D;gBACzF,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,2BAA2B;gBAC3B,OAAO;oBACL,IAAI,EAAE,IAAI,CAAC,IAAI;oBACf,SAAS,EAAE,IAAI,CAAC,SAAS;oBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;oBACrB,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,MAAM,EAAE,IAAI,CAAC,MAAM;oBACnB,WAAW,EAAE,IAAI,CAAC,WAAW;oBAC7B,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;oBAC/C,KAAK,EAAE,IAAI;oBACX,eAAe,EAAE,IAAI,CAAC,eAAe,IAAI,IAAI;oBAC7C,YAAY,EAAE,YAAY;oBAC1B,mBAAmB,EAAE;wBACnB,QAAQ,EAAE,IAAI,CAAC,SAAS;qBACzB;iBACF,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;gBAE1B,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;oBAC7B,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,IAAI,IAAI;oBAC/C,KAAK,EAAE,YAAY;oBACnB,eAAe,EAAE,IAAI;oBACrB,YAAY,EAAE,IAAI;oBAClB,mBAAmB,EAAE,EAAE;iBACxB,CAAC;YACJ,CAAC;QACH,CAAC;KAAA;IAED;;OAEG;IACH,IAAI,IAAI;QACN,OAAO,eAAe,CAAC;IACzB,CAAC;CACF"}
@@ -1,3 +0,0 @@
1
- export { HallucinationScorer } from './hallucination.js';
2
- export { HallucinationTemplate, HallucinationVerdictSchema, VerdictsSchema, ReasonSchema } from './prompts.js';
3
- //# sourceMappingURL=index.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../../src/scorers/metrics/hallucination/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,qBAAqB,EAAE,0BAA0B,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC"}