@arizeai/phoenix-evals 0.8.0 → 0.10.0

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 (248) hide show
  1. package/README.md +23 -23
  2. package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
  3. package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
  4. package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +58 -0
  5. package/dist/esm/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
  6. package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  7. package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +9 -1
  8. package/dist/esm/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  9. package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  10. package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js +14 -3
  11. package/dist/esm/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  12. package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  13. package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +16 -2
  14. package/dist/esm/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  15. package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  16. package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js +15 -1
  17. package/dist/esm/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  18. package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  19. package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -11
  20. package/dist/esm/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  21. package/dist/esm/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
  22. package/dist/esm/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
  23. package/dist/esm/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.js +86 -0
  24. package/dist/esm/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
  25. package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  26. package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js +44 -8
  27. package/dist/esm/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  28. package/dist/esm/__generated__/default_templates/index.d.ts +2 -0
  29. package/dist/esm/__generated__/default_templates/index.d.ts.map +1 -1
  30. package/dist/esm/__generated__/default_templates/index.js +2 -0
  31. package/dist/esm/__generated__/default_templates/index.js.map +1 -1
  32. package/dist/esm/core/EvaluatorBase.d.ts +2 -2
  33. package/dist/esm/core/EvaluatorBase.d.ts.map +1 -1
  34. package/dist/esm/core/FunctionEvaluator.d.ts +1 -1
  35. package/dist/esm/core/FunctionEvaluator.d.ts.map +1 -1
  36. package/dist/esm/core/FunctionEvaluator.js.map +1 -1
  37. package/dist/esm/helpers/asEvaluatorFn.d.ts +1 -1
  38. package/dist/esm/helpers/asEvaluatorFn.d.ts.map +1 -1
  39. package/dist/esm/helpers/asEvaluatorFn.js.map +1 -1
  40. package/dist/esm/helpers/createEvaluator.d.ts +2 -2
  41. package/dist/esm/helpers/createEvaluator.d.ts.map +1 -1
  42. package/dist/esm/helpers/createEvaluator.js.map +1 -1
  43. package/dist/esm/helpers/toEvaluationResult.d.ts +1 -1
  44. package/dist/esm/helpers/toEvaluationResult.d.ts.map +1 -1
  45. package/dist/esm/llm/ClassificationEvaluator.d.ts +3 -3
  46. package/dist/esm/llm/ClassificationEvaluator.d.ts.map +1 -1
  47. package/dist/esm/llm/ClassificationEvaluator.js.map +1 -1
  48. package/dist/esm/llm/LLMEvaluator.d.ts +1 -1
  49. package/dist/esm/llm/LLMEvaluator.d.ts.map +1 -1
  50. package/dist/esm/llm/createClassificationEvaluator.d.ts +1 -1
  51. package/dist/esm/llm/createClassificationEvaluator.d.ts.map +1 -1
  52. package/dist/esm/llm/createClassificationEvaluator.js.map +1 -1
  53. package/dist/esm/llm/createClassifierFn.d.ts +1 -1
  54. package/dist/esm/llm/createClassifierFn.d.ts.map +1 -1
  55. package/dist/esm/llm/createClassifierFn.js.map +1 -1
  56. package/dist/esm/llm/createConcisenessEvaluator.d.ts +43 -0
  57. package/dist/esm/llm/createConcisenessEvaluator.d.ts.map +1 -0
  58. package/dist/esm/llm/createConcisenessEvaluator.js +39 -0
  59. package/dist/esm/llm/createConcisenessEvaluator.js.map +1 -0
  60. package/dist/esm/llm/createCorrectnessEvaluator.d.ts +2 -2
  61. package/dist/esm/llm/createCorrectnessEvaluator.d.ts.map +1 -1
  62. package/dist/esm/llm/createCorrectnessEvaluator.js.map +1 -1
  63. package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts +2 -2
  64. package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
  65. package/dist/esm/llm/createDocumentRelevanceEvaluator.js.map +1 -1
  66. package/dist/esm/llm/createFaithfulnessEvaluator.d.ts +2 -2
  67. package/dist/esm/llm/createFaithfulnessEvaluator.d.ts.map +1 -1
  68. package/dist/esm/llm/createFaithfulnessEvaluator.js.map +1 -1
  69. package/dist/esm/llm/createHallucinationEvaluator.d.ts +2 -2
  70. package/dist/esm/llm/createHallucinationEvaluator.d.ts.map +1 -1
  71. package/dist/esm/llm/createHallucinationEvaluator.js.map +1 -1
  72. package/dist/esm/llm/createToolInvocationEvaluator.d.ts +2 -2
  73. package/dist/esm/llm/createToolInvocationEvaluator.d.ts.map +1 -1
  74. package/dist/esm/llm/createToolInvocationEvaluator.js.map +1 -1
  75. package/dist/esm/llm/createToolResponseHandlingEvaluator.d.ts +78 -0
  76. package/dist/esm/llm/createToolResponseHandlingEvaluator.d.ts.map +1 -0
  77. package/dist/esm/llm/createToolResponseHandlingEvaluator.js +59 -0
  78. package/dist/esm/llm/createToolResponseHandlingEvaluator.js.map +1 -0
  79. package/dist/esm/llm/createToolSelectionEvaluator.d.ts +64 -0
  80. package/dist/esm/llm/createToolSelectionEvaluator.d.ts.map +1 -0
  81. package/dist/esm/llm/createToolSelectionEvaluator.js +50 -0
  82. package/dist/esm/llm/createToolSelectionEvaluator.js.map +1 -0
  83. package/dist/esm/llm/generateClassification.d.ts +2 -2
  84. package/dist/esm/llm/generateClassification.d.ts.map +1 -1
  85. package/dist/esm/llm/generateClassification.js +1 -1
  86. package/dist/esm/llm/generateClassification.js.map +1 -1
  87. package/dist/esm/llm/index.d.ts +3 -0
  88. package/dist/esm/llm/index.d.ts.map +1 -1
  89. package/dist/esm/llm/index.js +3 -0
  90. package/dist/esm/llm/index.js.map +1 -1
  91. package/dist/esm/template/applyTemplate.d.ts +1 -1
  92. package/dist/esm/template/applyTemplate.d.ts.map +1 -1
  93. package/dist/esm/template/applyTemplate.js +1 -1
  94. package/dist/esm/template/applyTemplate.js.map +1 -1
  95. package/dist/esm/template/getTemplateVariables.d.ts +1 -1
  96. package/dist/esm/template/getTemplateVariables.d.ts.map +1 -1
  97. package/dist/esm/template/getTemplateVariables.js.map +1 -1
  98. package/dist/esm/tsconfig.esm.tsbuildinfo +1 -1
  99. package/dist/esm/types/evals.d.ts +4 -4
  100. package/dist/esm/types/evals.d.ts.map +1 -1
  101. package/dist/esm/types/otel.d.ts +1 -1
  102. package/dist/esm/types/otel.d.ts.map +1 -1
  103. package/dist/esm/utils/bindEvaluator.d.ts +2 -2
  104. package/dist/esm/utils/bindEvaluator.d.ts.map +1 -1
  105. package/dist/esm/utils/objectMappingUtils.d.ts +1 -1
  106. package/dist/esm/utils/objectMappingUtils.d.ts.map +1 -1
  107. package/dist/esm/utils/objectMappingUtils.js.map +1 -1
  108. package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
  109. package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
  110. package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -0
  111. package/dist/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
  112. package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  113. package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +9 -1
  114. package/dist/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  115. package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  116. package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js +14 -3
  117. package/dist/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  118. package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  119. package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js +16 -2
  120. package/dist/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  121. package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  122. package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js +15 -1
  123. package/dist/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  124. package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  125. package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js +61 -11
  126. package/dist/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  127. package/dist/src/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
  128. package/dist/src/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
  129. package/dist/src/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.js +89 -0
  130. package/dist/src/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
  131. package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -1
  132. package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js +44 -8
  133. package/dist/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -1
  134. package/dist/src/__generated__/default_templates/index.d.ts +2 -0
  135. package/dist/src/__generated__/default_templates/index.d.ts.map +1 -1
  136. package/dist/src/__generated__/default_templates/index.js +5 -1
  137. package/dist/src/__generated__/default_templates/index.js.map +1 -1
  138. package/dist/src/core/EvaluatorBase.d.ts +2 -2
  139. package/dist/src/core/EvaluatorBase.d.ts.map +1 -1
  140. package/dist/src/core/FunctionEvaluator.d.ts +1 -1
  141. package/dist/src/core/FunctionEvaluator.d.ts.map +1 -1
  142. package/dist/src/core/FunctionEvaluator.js.map +1 -1
  143. package/dist/src/helpers/asEvaluatorFn.d.ts +1 -1
  144. package/dist/src/helpers/asEvaluatorFn.d.ts.map +1 -1
  145. package/dist/src/helpers/asEvaluatorFn.js.map +1 -1
  146. package/dist/src/helpers/createEvaluator.d.ts +2 -2
  147. package/dist/src/helpers/createEvaluator.d.ts.map +1 -1
  148. package/dist/src/helpers/createEvaluator.js.map +1 -1
  149. package/dist/src/helpers/toEvaluationResult.d.ts +1 -1
  150. package/dist/src/helpers/toEvaluationResult.d.ts.map +1 -1
  151. package/dist/src/llm/ClassificationEvaluator.d.ts +3 -3
  152. package/dist/src/llm/ClassificationEvaluator.d.ts.map +1 -1
  153. package/dist/src/llm/ClassificationEvaluator.js.map +1 -1
  154. package/dist/src/llm/LLMEvaluator.d.ts +1 -1
  155. package/dist/src/llm/LLMEvaluator.d.ts.map +1 -1
  156. package/dist/src/llm/createClassificationEvaluator.d.ts +1 -1
  157. package/dist/src/llm/createClassificationEvaluator.d.ts.map +1 -1
  158. package/dist/src/llm/createClassificationEvaluator.js.map +1 -1
  159. package/dist/src/llm/createClassifierFn.d.ts +1 -1
  160. package/dist/src/llm/createClassifierFn.d.ts.map +1 -1
  161. package/dist/src/llm/createClassifierFn.js.map +1 -1
  162. package/dist/src/llm/createConcisenessEvaluator.d.ts +43 -0
  163. package/dist/src/llm/createConcisenessEvaluator.d.ts.map +1 -0
  164. package/dist/src/llm/createConcisenessEvaluator.js +50 -0
  165. package/dist/src/llm/createConcisenessEvaluator.js.map +1 -0
  166. package/dist/src/llm/createCorrectnessEvaluator.d.ts +2 -2
  167. package/dist/src/llm/createCorrectnessEvaluator.d.ts.map +1 -1
  168. package/dist/src/llm/createCorrectnessEvaluator.js.map +1 -1
  169. package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts +2 -2
  170. package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
  171. package/dist/src/llm/createDocumentRelevanceEvaluator.js.map +1 -1
  172. package/dist/src/llm/createFaithfulnessEvaluator.d.ts +2 -2
  173. package/dist/src/llm/createFaithfulnessEvaluator.d.ts.map +1 -1
  174. package/dist/src/llm/createFaithfulnessEvaluator.js.map +1 -1
  175. package/dist/src/llm/createHallucinationEvaluator.d.ts +2 -2
  176. package/dist/src/llm/createHallucinationEvaluator.d.ts.map +1 -1
  177. package/dist/src/llm/createHallucinationEvaluator.js.map +1 -1
  178. package/dist/src/llm/createToolInvocationEvaluator.d.ts +2 -2
  179. package/dist/src/llm/createToolInvocationEvaluator.d.ts.map +1 -1
  180. package/dist/src/llm/createToolInvocationEvaluator.js.map +1 -1
  181. package/dist/src/llm/createToolResponseHandlingEvaluator.d.ts +78 -0
  182. package/dist/src/llm/createToolResponseHandlingEvaluator.d.ts.map +1 -0
  183. package/dist/src/llm/createToolResponseHandlingEvaluator.js +70 -0
  184. package/dist/src/llm/createToolResponseHandlingEvaluator.js.map +1 -0
  185. package/dist/src/llm/createToolSelectionEvaluator.d.ts +64 -0
  186. package/dist/src/llm/createToolSelectionEvaluator.d.ts.map +1 -0
  187. package/dist/src/llm/createToolSelectionEvaluator.js +61 -0
  188. package/dist/src/llm/createToolSelectionEvaluator.js.map +1 -0
  189. package/dist/src/llm/generateClassification.d.ts +2 -2
  190. package/dist/src/llm/generateClassification.d.ts.map +1 -1
  191. package/dist/src/llm/generateClassification.js +1 -1
  192. package/dist/src/llm/generateClassification.js.map +1 -1
  193. package/dist/src/llm/index.d.ts +3 -0
  194. package/dist/src/llm/index.d.ts.map +1 -1
  195. package/dist/src/llm/index.js +3 -0
  196. package/dist/src/llm/index.js.map +1 -1
  197. package/dist/src/template/applyTemplate.d.ts +1 -1
  198. package/dist/src/template/applyTemplate.d.ts.map +1 -1
  199. package/dist/src/template/applyTemplate.js +1 -1
  200. package/dist/src/template/applyTemplate.js.map +1 -1
  201. package/dist/src/template/getTemplateVariables.d.ts +1 -1
  202. package/dist/src/template/getTemplateVariables.d.ts.map +1 -1
  203. package/dist/src/template/getTemplateVariables.js.map +1 -1
  204. package/dist/src/types/evals.d.ts +4 -4
  205. package/dist/src/types/evals.d.ts.map +1 -1
  206. package/dist/src/types/otel.d.ts +1 -1
  207. package/dist/src/types/otel.d.ts.map +1 -1
  208. package/dist/src/utils/bindEvaluator.d.ts +2 -2
  209. package/dist/src/utils/bindEvaluator.d.ts.map +1 -1
  210. package/dist/src/utils/objectMappingUtils.d.ts +1 -1
  211. package/dist/src/utils/objectMappingUtils.d.ts.map +1 -1
  212. package/dist/src/utils/objectMappingUtils.js.map +1 -1
  213. package/dist/tsconfig.tsbuildinfo +1 -1
  214. package/package.json +37 -38
  215. package/src/__generated__/default_templates/CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +60 -0
  216. package/src/__generated__/default_templates/CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +9 -1
  217. package/src/__generated__/default_templates/DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.ts +14 -3
  218. package/src/__generated__/default_templates/FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +16 -2
  219. package/src/__generated__/default_templates/HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG.ts +15 -1
  220. package/src/__generated__/default_templates/TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG.ts +61 -11
  221. package/src/__generated__/default_templates/TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG.ts +88 -0
  222. package/src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.ts +44 -8
  223. package/src/__generated__/default_templates/index.ts +2 -0
  224. package/src/core/EvaluatorBase.ts +2 -2
  225. package/src/core/FunctionEvaluator.ts +5 -2
  226. package/src/helpers/asEvaluatorFn.ts +1 -2
  227. package/src/helpers/createEvaluator.ts +2 -3
  228. package/src/helpers/toEvaluationResult.ts +1 -1
  229. package/src/llm/ClassificationEvaluator.ts +4 -5
  230. package/src/llm/LLMEvaluator.ts +1 -1
  231. package/src/llm/createClassificationEvaluator.ts +1 -2
  232. package/src/llm/createClassifierFn.ts +1 -2
  233. package/src/llm/createConcisenessEvaluator.ts +71 -0
  234. package/src/llm/createCorrectnessEvaluator.ts +2 -3
  235. package/src/llm/createDocumentRelevanceEvaluator.ts +2 -3
  236. package/src/llm/createFaithfulnessEvaluator.ts +2 -3
  237. package/src/llm/createHallucinationEvaluator.ts +2 -3
  238. package/src/llm/createToolInvocationEvaluator.ts +2 -3
  239. package/src/llm/createToolResponseHandlingEvaluator.ts +108 -0
  240. package/src/llm/createToolSelectionEvaluator.ts +92 -0
  241. package/src/llm/generateClassification.ts +5 -5
  242. package/src/llm/index.ts +3 -0
  243. package/src/template/applyTemplate.ts +2 -3
  244. package/src/template/getTemplateVariables.ts +2 -2
  245. package/src/types/evals.ts +4 -4
  246. package/src/types/otel.ts +1 -1
  247. package/src/utils/bindEvaluator.ts +2 -2
  248. package/src/utils/objectMappingUtils.ts +2 -2
@@ -4,20 +4,56 @@ Object.defineProperty(exports, "__esModule", { value: true });
4
4
  exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = void 0;
5
5
  exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = {
6
6
  name: "tool_selection",
7
- description: "For determining if the correct tool was selected for a given context. Requires conversation context, a list of available tools, and the LLM's tool selections.",
7
+ description: "For determining if the correct tool was selected for a given context.",
8
8
  optimizationDirection: "MAXIMIZE",
9
9
  template: [
10
10
  {
11
11
  role: "user",
12
12
  content: `
13
13
  You are an impartial judge evaluating an LLM's tool-calling behavior, specifically whether the LLM selected the most appropriate tool or tools for the task.
14
- Your task: Determine whether the LLM's tool selection was correct or incorrect based on: - The conversation context - The available tools - The LLM's tool invocation(s)
15
- Criteria Return "correct" only when ALL of the following are true: - The LLM chose the best available tool for the user query OR correctly avoided tools if none were needed. - The tool name exists in the available tools list. - The tool is allowed and safe to call. - The LLM selected the correct number of tools for the task.
16
- Return "incorrect" if ANY of the following are true: - The LLM used a hallucinated or nonexistent tool. - The LLM selected a tool when none was needed. - The LLM did not use a tool when one was required. - The LLM chose a suboptimal or irrelevant tool. - The LLM selected an unsafe or not-permitted tool. - The tool name does not appear in the available tools list.
17
- Before providing your final judgment, explain your reasoning and consider: - What does the input context require? - Can this be answered without tools, or is a tool necessary? - If a tool was selected, does it exist in the available tools? - Does the selected tool's description match the user's needs? - Is the selection safe and appropriate? - Is there a better tool available that should have been chosen instead?
18
- <data> <context> {{input}} </context>
19
- <available_tools> {{availableTools}} </available_tools>
20
- <tool_selection> {{toolSelection}} </tool_selection> </data>
14
+
15
+ Your task: Determine whether the LLM's tool selection was correct or incorrect based on:
16
+ - The conversation context (input)
17
+ - The available tools
18
+ - The LLM's output and tool invocation(s)
19
+
20
+ Criteria
21
+ Return "correct" only when ALL of the following are true:
22
+ - The LLM chose the best available tool for the user query OR correctly avoided tools if none were needed.
23
+ - The tool name exists in the available tools list.
24
+ - The tool is allowed and safe to call.
25
+ - The LLM selected the correct number of tools for the task.
26
+
27
+ Return "incorrect" if ANY of the following are true:
28
+ - The LLM used a hallucinated or nonexistent tool.
29
+ - The LLM selected a tool when none was needed.
30
+ - The LLM did not use a tool when one was required.
31
+ - The LLM chose a suboptimal or irrelevant tool.
32
+ - The LLM selected an unsafe or not-permitted tool.
33
+ - The tool name does not appear in the available tools list.
34
+
35
+ Before providing your final judgment, explain your reasoning and consider:
36
+ - What does the input context require?
37
+ - Can this be answered without tools, or is a tool necessary?
38
+ - If a tool was selected, does it exist in the available tools?
39
+ - Does the selected tool's description match the user's needs?
40
+ - Is the selection safe and appropriate?
41
+ - Is there a better tool available that should have been chosen instead?
42
+
43
+ <data>
44
+ <input>
45
+ {{input}}
46
+ </input>
47
+
48
+ <available_tools>
49
+ {{availableTools}}
50
+ </available_tools>
51
+
52
+ <output>
53
+ {{toolSelection}}
54
+ </output>
55
+ </data>
56
+
21
57
  Given the above data, is the tool selection correct or incorrect?
22
58
  `,
23
59
  },
@@ -1 +1 @@
1
- {"version":3,"file":"TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAIlC,QAAA,8CAA8C,GAAkC;IAC3F,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,gKAAgK;IAC7K,qBAAqB,EAAE,UAAU;IACjC,QAAQ,EAAE;QACR;YACE,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE;;;;;;;;;;CAUd;SACI;KACF;IACD,OAAO,EAAE;QACT,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,CAAC;KACf;CACA,CAAC"}
1
+ {"version":3,"file":"TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAIlC,QAAA,8CAA8C,GAAkC;IAC3F,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,uEAAuE;IACpF,qBAAqB,EAAE,UAAU;IACjC,QAAQ,EAAE;QACR;YACE,IAAI,EAAE,MAAM;YACZ,OAAO,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8Cd;SACI;KACF;IACD,OAAO,EAAE;QACT,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,CAAC;KACf;CACA,CAAC"}
@@ -1,7 +1,9 @@
1
+ export { CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG";
1
2
  export { CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG";
2
3
  export { DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG } from "./DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG";
3
4
  export { FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG } from "./FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG";
4
5
  export { HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG } from "./HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG";
5
6
  export { TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG";
7
+ export { TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG";
6
8
  export { TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG } from "./TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG";
7
9
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,kDAAkD,EAAE,MAAM,sDAAsD,CAAC;AAC1H,OAAO,EAAE,4CAA4C,EAAE,MAAM,gDAAgD,CAAC;AAC9G,OAAO,EAAE,6CAA6C,EAAE,MAAM,iDAAiD,CAAC;AAChH,OAAO,EAAE,+CAA+C,EAAE,MAAM,mDAAmD,CAAC;AACpH,OAAO,EAAE,8CAA8C,EAAE,MAAM,kDAAkD,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,2CAA2C,EAAE,MAAM,+CAA+C,CAAC;AAC5G,OAAO,EAAE,kDAAkD,EAAE,MAAM,sDAAsD,CAAC;AAC1H,OAAO,EAAE,4CAA4C,EAAE,MAAM,gDAAgD,CAAC;AAC9G,OAAO,EAAE,6CAA6C,EAAE,MAAM,iDAAiD,CAAC;AAChH,OAAO,EAAE,+CAA+C,EAAE,MAAM,mDAAmD,CAAC;AACpH,OAAO,EAAE,sDAAsD,EAAE,MAAM,0DAA0D,CAAC;AAClI,OAAO,EAAE,8CAA8C,EAAE,MAAM,kDAAkD,CAAC"}
@@ -1,7 +1,9 @@
1
1
  "use strict";
2
2
  // This file is generated. Do not edit by hand.
3
3
  Object.defineProperty(exports, "__esModule", { value: true });
4
- exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG = exports.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG = void 0;
4
+ exports.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG = exports.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG = exports.FAITHFULNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG = exports.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG = exports.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG = void 0;
5
+ var CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG");
6
+ Object.defineProperty(exports, "CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG; } });
5
7
  var CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG");
6
8
  Object.defineProperty(exports, "CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG_1.CORRECTNESS_CLASSIFICATION_EVALUATOR_CONFIG; } });
7
9
  var DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG");
@@ -12,6 +14,8 @@ var HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./HALLUCINATION_C
12
14
  Object.defineProperty(exports, "HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG_1.HALLUCINATION_CLASSIFICATION_EVALUATOR_CONFIG; } });
13
15
  var TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG");
14
16
  Object.defineProperty(exports, "TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG_1.TOOL_INVOCATION_CLASSIFICATION_EVALUATOR_CONFIG; } });
17
+ var TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG");
18
+ Object.defineProperty(exports, "TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG_1.TOOL_RESPONSE_HANDLING_CLASSIFICATION_EVALUATOR_CONFIG; } });
15
19
  var TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG_1 = require("./TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG");
16
20
  Object.defineProperty(exports, "TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG", { enumerable: true, get: function () { return TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG_1.TOOL_SELECTION_CLASSIFICATION_EVALUATOR_CONFIG; } });
17
21
  //# sourceMappingURL=index.js.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAE/C,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,2HAA0H;AAAjH,wLAAA,kDAAkD,OAAA;AAC3D,+GAA8G;AAArG,4KAAA,4CAA4C,OAAA;AACrD,iHAAgH;AAAvG,8KAAA,6CAA6C,OAAA;AACtD,qHAAoH;AAA3G,kLAAA,+CAA+C,OAAA;AACxD,mHAAkH;AAAzG,gLAAA,8CAA8C,OAAA"}
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/__generated__/default_templates/index.ts"],"names":[],"mappings":";AAAA,+CAA+C;;;AAE/C,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,6GAA4G;AAAnG,0KAAA,2CAA2C,OAAA;AACpD,2HAA0H;AAAjH,wLAAA,kDAAkD,OAAA;AAC3D,+GAA8G;AAArG,4KAAA,4CAA4C,OAAA;AACrD,iHAAgH;AAAvG,8KAAA,6CAA6C,OAAA;AACtD,qHAAoH;AAA3G,kLAAA,+CAA+C,OAAA;AACxD,mIAAkI;AAAzH,gMAAA,sDAAsD,OAAA;AAC/D,mHAAkH;AAAzG,gLAAA,8CAA8C,OAAA"}
@@ -1,5 +1,5 @@
1
- import { CreateEvaluatorArgs, EvaluationKind, EvaluationResult, EvaluatorInterface, OptimizationDirection, TelemetryConfig } from "../types";
2
- import { ObjectMapping } from "../types/data";
1
+ import type { CreateEvaluatorArgs, EvaluationKind, EvaluationResult, EvaluatorInterface, OptimizationDirection, TelemetryConfig } from "../types";
2
+ import type { ObjectMapping } from "../types/data";
3
3
  /**
4
4
  * Base class for all evaluators
5
5
  */
@@ -1 +1 @@
1
- {"version":3,"file":"EvaluatorBase.d.ts","sourceRoot":"","sources":["../../../src/core/EvaluatorBase.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,mBAAmB,EACnB,cAAc,EACd,gBAAgB,EAChB,kBAAkB,EAClB,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAE9C;;GAEG;AACH,8BAAsB,aAAa,CACjC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,YAAW,kBAAkB,CAAC,UAAU,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,cAAc,CAAC;IAC9B,QAAQ,CAAC,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IACvD,QAAQ,CAAC,YAAY,CAAC,EAAE,aAAa,CAAC,UAAU,CAAC,CAAC;IAClD,QAAQ,CAAC,SAAS,CAAC,EAAE,eAAe,CAAC;gBACzB,EACV,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,YAAY,EACZ,SAAS,GACV,EAAE,mBAAmB,CAAC,UAAU,CAAC;IAOlC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAElE;;OAEG;IACH,QAAQ,CAAC,gBAAgB,CACvB,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,aAAa,CAAC,UAAU,CAAC;CAC7B"}
1
+ {"version":3,"file":"EvaluatorBase.d.ts","sourceRoot":"","sources":["../../../src/core/EvaluatorBase.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,mBAAmB,EACnB,cAAc,EACd,gBAAgB,EAChB,kBAAkB,EAClB,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAClB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAEnD;;GAEG;AACH,8BAAsB,aAAa,CACjC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,YAAW,kBAAkB,CAAC,UAAU,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,IAAI,EAAE,cAAc,CAAC;IAC9B,QAAQ,CAAC,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IACvD,QAAQ,CAAC,YAAY,CAAC,EAAE,aAAa,CAAC,UAAU,CAAC,CAAC;IAClD,QAAQ,CAAC,SAAS,CAAC,EAAE,eAAe,CAAC;gBACzB,EACV,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,YAAY,EACZ,SAAS,GACV,EAAE,mBAAmB,CAAC,UAAU,CAAC;IAOlC,QAAQ,CAAC,QAAQ,CAAC,QAAQ,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAElE;;OAEG;IACH,QAAQ,CAAC,gBAAgB,CACvB,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,aAAa,CAAC,UAAU,CAAC;CAC7B"}
@@ -1,4 +1,4 @@
1
- import { CreateEvaluatorArgs, EvaluationResult, ObjectMapping } from "../types";
1
+ import type { CreateEvaluatorArgs, EvaluationResult, ObjectMapping } from "../types";
2
2
  import { EvaluatorBase } from "./EvaluatorBase";
3
3
  type FunctionEvaluatorArgs<RecordType extends Record<string, unknown>> = CreateEvaluatorArgs<RecordType> & {
4
4
  evaluateFn: (args: RecordType) => Promise<EvaluationResult>;
@@ -1 +1 @@
1
- {"version":3,"file":"FunctionEvaluator.d.ts","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEhF,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAEhD,KAAK,qBAAqB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IACnE,mBAAmB,CAAC,UAAU,CAAC,GAAG;IAChC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;CAC7D,CAAC;AACJ;;GAEG;AACH,qBAAa,iBAAiB,CAC5B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;IACjC,QAAQ,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;gBACzD,EAAE,UAAU,EAAE,GAAG,IAAI,EAAE,EAAE,qBAAqB,CAAC,UAAU,CAAC;IAIhE,QAAQ,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAG3D,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,iBAAiB,CAAC,UAAU,CAAC;CAGjC"}
1
+ {"version":3,"file":"FunctionEvaluator.d.ts","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,mBAAmB,EACnB,gBAAgB,EAChB,aAAa,EACd,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAEhD,KAAK,qBAAqB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IACnE,mBAAmB,CAAC,UAAU,CAAC,GAAG;IAChC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;CAC7D,CAAC;AACJ;;GAEG;AACH,qBAAa,iBAAiB,CAC5B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;IACjC,QAAQ,CAAC,UAAU,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,OAAO,CAAC,gBAAgB,CAAC,CAAC;gBACzD,EAAE,UAAU,EAAE,GAAG,IAAI,EAAE,EAAE,qBAAqB,CAAC,UAAU,CAAC;IAIhE,QAAQ,CAAC,IAAI,EAAE,UAAU,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAG3D,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,iBAAiB,CAAC,UAAU,CAAC;CAGjC"}
@@ -1 +1 @@
1
- {"version":3,"file":"FunctionEvaluator.js","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;;AAEA,mDAAgD;AAMhD;;GAEG;AACH,MAAa,iBAEX,SAAQ,6BAAyB;IAEjC,YAAY,EAA0D;YAA1D,EAAE,UAAU,OAA8C,EAAzC,IAAI,cAArB,cAAuB,CAAF;QAC/B,KAAK,mBAAM,IAAI,EAAG,CAAC;QACnB,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;IAC/B,CAAC;IACD,KAAK,CAAC,QAAQ,CAAC,IAAgB;QAC7B,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IACD,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,iBAAiB,iCAAM,IAAI,KAAE,YAAY,IAAG,CAAC;IAC1D,CAAC;CACF;AAhBD,8CAgBC"}
1
+ {"version":3,"file":"FunctionEvaluator.js","sourceRoot":"","sources":["../../../src/core/FunctionEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;;AAKA,mDAAgD;AAMhD;;GAEG;AACH,MAAa,iBAEX,SAAQ,6BAAyB;IAEjC,YAAY,EAA0D;YAA1D,EAAE,UAAU,OAA8C,EAAzC,IAAI,cAArB,cAAuB,CAAF;QAC/B,KAAK,mBAAM,IAAI,EAAG,CAAC;QACnB,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;IAC/B,CAAC;IACD,KAAK,CAAC,QAAQ,CAAC,IAAgB;QAC7B,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IACD,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,iBAAiB,iCAAM,IAAI,KAAE,YAAY,IAAG,CAAC;IAC1D,CAAC;CACF;AAhBD,8CAgBC"}
@@ -1,4 +1,4 @@
1
- import { AnyFn, EvaluatorFn } from "../types";
1
+ import type { AnyFn, EvaluatorFn } from "../types";
2
2
  /**
3
3
  * A function that converts a generic function into an evaluator function
4
4
  */
@@ -1 +1 @@
1
- {"version":3,"file":"asEvaluatorFn.d.ts","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAK9C;;GAEG;AACH,wBAAgB,aAAa,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACtE,EAAE,EAAE,KAAK,GACR,WAAW,CAAC,UAAU,CAAC,CAQzB"}
1
+ {"version":3,"file":"asEvaluatorFn.d.ts","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAInD;;GAEG;AACH,wBAAgB,aAAa,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACtE,EAAE,EAAE,KAAK,GACR,WAAW,CAAC,UAAU,CAAC,CAQzB"}
@@ -1 +1 @@
1
- {"version":3,"file":"asEvaluatorFn.js","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":";;AAQA,sCAUC;AAjBD,kDAA+C;AAE/C,6DAA0D;AAE1D;;GAEG;AACH,SAAgB,aAAa,CAC3B,EAAS;IAET,OAAO,KAAK,EAAE,GAAG,IAAI,EAAE,EAAE;QACvB,IAAI,MAAM,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACzB,IAAI,IAAA,qBAAS,EAAC,MAAM,CAAC,EAAE,CAAC;YACtB,MAAM,GAAG,MAAM,MAAM,CAAC;QACxB,CAAC;QACD,OAAO,IAAA,uCAAkB,EAAC,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"asEvaluatorFn.js","sourceRoot":"","sources":["../../../src/helpers/asEvaluatorFn.ts"],"names":[],"mappings":";;AAOA,sCAUC;AAhBD,kDAA+C;AAC/C,6DAA0D;AAE1D;;GAEG;AACH,SAAgB,aAAa,CAC3B,EAAS;IAET,OAAO,KAAK,EAAE,GAAG,IAAI,EAAE,EAAE;QACvB,IAAI,MAAM,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACzB,IAAI,IAAA,qBAAS,EAAC,MAAM,CAAC,EAAE,CAAC;YACtB,MAAM,GAAG,MAAM,MAAM,CAAC;QACxB,CAAC;QACD,OAAO,IAAA,uCAAkB,EAAC,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC;AACJ,CAAC"}
@@ -1,5 +1,5 @@
1
- import { EvaluatorBase } from "../core/EvaluatorBase";
2
- import { EvaluationKind, OptimizationDirection, TelemetryConfig } from "../types";
1
+ import type { EvaluatorBase } from "../core/EvaluatorBase";
2
+ import type { EvaluationKind, OptimizationDirection, TelemetryConfig } from "../types";
3
3
  type AnyFn = (...args: any[]) => any;
4
4
  /**
5
5
  * Options for creating a custom evaluator using {@link createEvaluator}.
@@ -1 +1 @@
1
- {"version":3,"file":"createEvaluator.d.ts","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAEtD,OAAO,EACL,cAAc,EACd,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAKlB,KAAK,KAAK,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,KAAK,GAAG,CAAC;AAMrC;;;;GAIG;AACH,MAAM,MAAM,sBAAsB,GAAG;IACnC;;;;;;;;;;OAUG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;;;;;;;;;OAYG;IACH,IAAI,CAAC,EAAE,cAAc,CAAC;IACtB;;;;;;;;;;;;;;OAcG;IACH,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IAC9C;;;;;;;;;;;;;;;OAeG;IACH,SAAS,CAAC,EAAE,eAAe,CAAC;CAC7B,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,wBAAgB,eAAe,CAC7B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACpE,EAAE,SAAS,KAAK,GAAG,KAAK,EACxB,EAAE,EAAE,EAAE,EAAE,OAAO,CAAC,EAAE,sBAAsB,GAAG,aAAa,CAAC,UAAU,CAAC,CAyBrE"}
1
+ {"version":3,"file":"createEvaluator.d.ts","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAE3D,OAAO,KAAK,EACV,cAAc,EACd,qBAAqB,EACrB,eAAe,EAChB,MAAM,UAAU,CAAC;AAIlB,KAAK,KAAK,GAAG,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,KAAK,GAAG,CAAC;AAMrC;;;;GAIG;AACH,MAAM,MAAM,sBAAsB,GAAG;IACnC;;;;;;;;;;OAUG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;;;;;;;;;OAYG;IACH,IAAI,CAAC,EAAE,cAAc,CAAC;IACtB;;;;;;;;;;;;;;OAcG;IACH,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IAC9C;;;;;;;;;;;;;;;OAeG;IACH,SAAS,CAAC,EAAE,eAAe,CAAC;CAC7B,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,wBAAgB,eAAe,CAC7B,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACpE,EAAE,SAAS,KAAK,GAAG,KAAK,EACxB,EAAE,EAAE,EAAE,EAAE,OAAO,CAAC,EAAE,sBAAsB,GAAG,aAAa,CAAC,UAAU,CAAC,CAyBrE"}
@@ -1 +1 @@
1
- {"version":3,"file":"createEvaluator.js","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":";;AA2JA,0CA4BC;AAvLD,oEAAuD;AAGvD,iEAA8D;AAO9D,mDAAgD;AAKhD,SAAS,kBAAkB;IACzB,OAAO,aAAa,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;AACpE,CAAC;AAqED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,SAAgB,eAAe,CAG7B,EAAM,EAAE,OAAgC;IACxC,MAAM,EACJ,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,SAAS,GAAG,EAAE,SAAS,EAAE,IAAI,EAAE,GAChC,GAAG,OAAO,IAAI,EAAE,CAAC;IAClB,MAAM,aAAa,GAAG,IAAI,IAAI,EAAE,CAAC,IAAI,IAAI,kBAAkB,EAAE,CAAC;IAC9D,IAAI,UAAU,GAAG,IAAA,6BAAa,EAAa,EAAE,CAAC,CAAC;IAE/C,0DAA0D;IAC1D,IAAI,SAAS,IAAI,SAAS,CAAC,SAAS,EAAE,CAAC;QACrC,UAAU,GAAG,IAAA,6BAAQ,EAAC,UAAU,EAAE;YAChC,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,IAAI,EAAE,aAAa;YACnB,IAAI,EAAE,WAAW;SAClB,CAAC,CAAC;IACL,CAAC;IACD,OAAO,IAAI,qCAAiB,CAAa;QACvC,UAAU;QACV,IAAI,EAAE,aAAa;QACnB,IAAI,EAAE,IAAI,IAAI,MAAM;QACpB,qBAAqB,EAAE,qBAAqB,IAAI,UAAU;QAC1D,SAAS;KACV,CAAC,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createEvaluator.js","sourceRoot":"","sources":["../../../src/helpers/createEvaluator.ts"],"names":[],"mappings":";;AA0JA,0CA4BC;AAtLD,oEAAuD;AAGvD,iEAA8D;AAM9D,mDAAgD;AAKhD,SAAS,kBAAkB;IACzB,OAAO,aAAa,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,SAAS,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;AACpE,CAAC;AAqED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoEG;AACH,SAAgB,eAAe,CAG7B,EAAM,EAAE,OAAgC;IACxC,MAAM,EACJ,IAAI,EACJ,IAAI,EACJ,qBAAqB,EACrB,SAAS,GAAG,EAAE,SAAS,EAAE,IAAI,EAAE,GAChC,GAAG,OAAO,IAAI,EAAE,CAAC;IAClB,MAAM,aAAa,GAAG,IAAI,IAAI,EAAE,CAAC,IAAI,IAAI,kBAAkB,EAAE,CAAC;IAC9D,IAAI,UAAU,GAAG,IAAA,6BAAa,EAAa,EAAE,CAAC,CAAC;IAE/C,0DAA0D;IAC1D,IAAI,SAAS,IAAI,SAAS,CAAC,SAAS,EAAE,CAAC;QACrC,UAAU,GAAG,IAAA,6BAAQ,EAAC,UAAU,EAAE;YAChC,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,IAAI,EAAE,aAAa;YACnB,IAAI,EAAE,WAAW;SAClB,CAAC,CAAC;IACL,CAAC;IACD,OAAO,IAAI,qCAAiB,CAAa;QACvC,UAAU;QACV,IAAI,EAAE,aAAa;QACnB,IAAI,EAAE,IAAI,IAAI,MAAM;QACpB,qBAAqB,EAAE,qBAAqB,IAAI,UAAU;QAC1D,SAAS;KACV,CAAC,CAAC;AACL,CAAC"}
@@ -1,4 +1,4 @@
1
- import { EvaluationResult } from "../types";
1
+ import type { EvaluationResult } from "../types";
2
2
  /**
3
3
  * Converts an unknown value to an {@link EvaluationResult}.
4
4
  *
@@ -1 +1 @@
1
- {"version":3,"file":"toEvaluationResult.d.ts","sourceRoot":"","sources":["../../../src/helpers/toEvaluationResult.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,MAAM,UAAU,CAAC;AAuD5C;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+DG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,OAAO,GAAG,gBAAgB,CAyBpE"}
1
+ {"version":3,"file":"toEvaluationResult.d.ts","sourceRoot":"","sources":["../../../src/helpers/toEvaluationResult.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,UAAU,CAAC;AAuDjD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+DG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,OAAO,GAAG,gBAAgB,CAyBpE"}
@@ -1,7 +1,7 @@
1
- import { ClassificationChoicesMap, CreateClassificationEvaluatorArgs, EvaluatorFn, PromptTemplate, WithPromptTemplate } from "../types";
2
- import { ObjectMapping } from "../types/data";
1
+ import type { LanguageModel } from "ai";
2
+ import type { ClassificationChoicesMap, CreateClassificationEvaluatorArgs, EvaluatorFn, PromptTemplate, WithPromptTemplate } from "../types";
3
+ import type { ObjectMapping } from "../types/data";
3
4
  import { LLMEvaluator } from "./LLMEvaluator";
4
- import { LanguageModel } from "ai";
5
5
  /**
6
6
  * An LLM evaluator that performs evaluation via classification
7
7
  */
@@ -1 +1 @@
1
- {"version":3,"file":"ClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EACL,wBAAwB,EACxB,iCAAiC,EACjC,WAAW,EACX,cAAc,EACd,kBAAkB,EACnB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAI9C,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C,OAAO,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAEnC;;GAEG;AACH,qBAAa,uBAAuB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC7E,SAAQ,YAAY,CAAC,UAAU,CAC/B,YAAW,kBAAkB;IAE7B,QAAQ,CAAC,WAAW,EAAE,WAAW,CAAC,UAAU,CAAC,CAAC;IAC9C,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC;IACxC;;OAEG;IACH,OAAO,CAAC,wBAAwB,CAAuB;IACvD;;OAEG;IACH,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC;IAC9B;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,wBAAwB,CAAC;gBAE/B,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC;IAS/D,QAAQ,GAAI,SAAS,UAAU,kDAM7B;IACF;;OAEG;IACH,IAAI,uBAAuB,IAAI,MAAM,EAAE,CAStC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,uBAAuB,CAAC,UAAU,CAAC;CAMvC"}
1
+ {"version":3,"file":"ClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAGxC,OAAO,KAAK,EACV,wBAAwB,EACxB,iCAAiC,EACjC,WAAW,EACX,cAAc,EACd,kBAAkB,EACnB,MAAM,UAAU,CAAC;AAClB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAGnD,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAE9C;;GAEG;AACH,qBAAa,uBAAuB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC7E,SAAQ,YAAY,CAAC,UAAU,CAC/B,YAAW,kBAAkB;IAE7B,QAAQ,CAAC,WAAW,EAAE,WAAW,CAAC,UAAU,CAAC,CAAC;IAC9C,QAAQ,CAAC,cAAc,EAAE,cAAc,CAAC;IACxC;;OAEG;IACH,OAAO,CAAC,wBAAwB,CAAuB;IACvD;;OAEG;IACH,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC;IAC9B;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,wBAAwB,CAAC;gBAE/B,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC;IAS/D,QAAQ,GAAI,SAAS,UAAU,kDAM7B;IACF;;OAEG;IACH,IAAI,uBAAuB,IAAI,MAAM,EAAE,CAStC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,GACtC,uBAAuB,CAAC,UAAU,CAAC;CAMvC"}
@@ -1 +1 @@
1
- {"version":3,"file":"ClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":";;;AAAA,0CAAmD;AASnD,oEAA0D;AAE1D,6DAA0D;AAC1D,iDAA8C;AAI9C;;GAEG;AACH,MAAa,uBACX,SAAQ,2BAAwB;IAkBhC,YAAY,IAAmD;QAC7D,KAAK,CAAC,IAAI,CAAC,CAAC;QAQd,aAAQ,GAAG,CAAC,OAAmB,EAAE,EAAE;YACjC,OAAO,IAAI,CAAC,WAAW,CACrB,IAAI,CAAC,YAAY;gBACf,CAAC,CAAC,IAAA,gCAAW,EAAa,OAAO,EAAE,IAAI,CAAC,YAAY,CAAC;gBACrD,CAAC,CAAC,OAAO,CACZ,CAAC;QACJ,CAAC,CAAC;QAbA,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,cAAc,CAAC;QAC1C,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACxB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;QAC5B,IAAI,CAAC,WAAW,GAAG,IAAA,uCAAkB,oBAChC,IAAI,EACP,CAAC;IACL,CAAC;IAQD;;OAEG;IACH,IAAI,uBAAuB;QACzB,0DAA0D;QAC1D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,wBAAwB,CAAC,EAAE,CAAC;YAClD,IAAI,CAAC,wBAAwB,GAAG,IAAA,+BAAoB,EAAC;gBACnD,QAAQ,EAAE,IAAI,CAAC,cAAc;aAC9B,CAAC,CAAC;QACL,CAAC;QACD,+BAA+B;QAC/B,OAAO,CAAC,GAAG,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC5C,CAAC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,uBAAuB,iCAC7B,IAAI,KACP,YAAY,IACZ,CAAC;IACL,CAAC;CACF;AA3DD,0DA2DC"}
1
+ {"version":3,"file":"ClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/ClassificationEvaluator.ts"],"names":[],"mappings":";;;AAEA,0CAAmD;AASnD,oEAA0D;AAC1D,6DAA0D;AAC1D,iDAA8C;AAE9C;;GAEG;AACH,MAAa,uBACX,SAAQ,2BAAwB;IAkBhC,YAAY,IAAmD;QAC7D,KAAK,CAAC,IAAI,CAAC,CAAC;QAQd,aAAQ,GAAG,CAAC,OAAmB,EAAE,EAAE;YACjC,OAAO,IAAI,CAAC,WAAW,CACrB,IAAI,CAAC,YAAY;gBACf,CAAC,CAAC,IAAA,gCAAW,EAAa,OAAO,EAAE,IAAI,CAAC,YAAY,CAAC;gBACrD,CAAC,CAAC,OAAO,CACZ,CAAC;QACJ,CAAC,CAAC;QAbA,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,cAAc,CAAC;QAC1C,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACxB,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;QAC5B,IAAI,CAAC,WAAW,GAAG,IAAA,uCAAkB,oBAChC,IAAI,EACP,CAAC;IACL,CAAC;IAQD;;OAEG;IACH,IAAI,uBAAuB;QACzB,0DAA0D;QAC1D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,wBAAwB,CAAC,EAAE,CAAC;YAClD,IAAI,CAAC,wBAAwB,GAAG,IAAA,+BAAoB,EAAC;gBACnD,QAAQ,EAAE,IAAI,CAAC,cAAc;aAC9B,CAAC,CAAC;QACL,CAAC;QACD,+BAA+B;QAC/B,OAAO,CAAC,GAAG,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC5C,CAAC;IACD;;OAEG;IACH,gBAAgB,CACd,YAAuC;QAEvC,OAAO,IAAI,uBAAuB,iCAC7B,IAAI,KACP,YAAY,IACZ,CAAC;IACL,CAAC;CACF;AA3DD,0DA2DC"}
@@ -1,5 +1,5 @@
1
1
  import { EvaluatorBase } from "../core/EvaluatorBase";
2
- import { CreateLLMEvaluatorArgs } from "../types";
2
+ import type { CreateLLMEvaluatorArgs } from "../types";
3
3
  /**
4
4
  * Base class for llm evaluation metrics / scores
5
5
  */
@@ -1 +1 @@
1
- {"version":3,"file":"LLMEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/LLMEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AACtD,OAAO,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAElD;;GAEG;AACH,8BAAsB,YAAY,CAChC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;gBACrB,EAAE,GAAG,IAAI,EAAE,EAAE,sBAAsB,CAAC,UAAU,CAAC;CAG5D"}
1
+ {"version":3,"file":"LLMEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/LLMEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AACtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,UAAU,CAAC;AAEvD;;GAEG;AACH,8BAAsB,YAAY,CAChC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC1C,SAAQ,aAAa,CAAC,UAAU,CAAC;gBACrB,EAAE,GAAG,IAAI,EAAE,EAAE,sBAAsB,CAAC,UAAU,CAAC;CAG5D"}
@@ -1,4 +1,4 @@
1
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
2
  import { ClassificationEvaluator } from "./ClassificationEvaluator";
3
3
  export declare function createClassificationEvaluator<RecordType extends Record<string, unknown>>(args: CreateClassificationEvaluatorArgs<RecordType>): ClassificationEvaluator<RecordType>;
4
4
  //# sourceMappingURL=createClassificationEvaluator.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"createClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAEpE,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAE1C,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC,GAClD,uBAAuB,CAAC,UAAU,CAAC,CAErC"}
1
+ {"version":3,"file":"createClassificationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAEpE,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAE1C,IAAI,EAAE,iCAAiC,CAAC,UAAU,CAAC,GAClD,uBAAuB,CAAC,UAAU,CAAC,CAErC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":";;AAIA,sEAMC;AARD,uEAAoE;AAEpE,SAAgB,6BAA6B,CAG3C,IAAmD;IAEnD,OAAO,IAAI,iDAAuB,CAAa,IAAI,CAAC,CAAC;AACvD,CAAC"}
1
+ {"version":3,"file":"createClassificationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createClassificationEvaluator.ts"],"names":[],"mappings":";;AAGA,sEAMC;AARD,uEAAoE;AAEpE,SAAgB,6BAA6B,CAG3C,IAAmD;IAEnD,OAAO,IAAI,iDAAuB,CAAa,IAAI,CAAC,CAAC;AACvD,CAAC"}
@@ -1,4 +1,4 @@
1
- import { CreateClassifierArgs, EvaluatorFn } from "../types/evals";
1
+ import type { CreateClassifierArgs, EvaluatorFn } from "../types/evals";
2
2
  /**
3
3
  * A function that serves as a factory that will output a classification evaluator function
4
4
  */
@@ -1 +1 @@
1
- {"version":3,"file":"createClassifierFn.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,oBAAoB,EAEpB,WAAW,EACZ,MAAM,gBAAgB,CAAC;AAkBxB;;GAEG;AACH,wBAAgB,kBAAkB,CAChC,gBAAgB,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAChD,IAAI,EAAE,oBAAoB,GAAG,WAAW,CAAC,gBAAgB,CAAC,CA4B3D"}
1
+ {"version":3,"file":"createClassifierFn.d.ts","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAEV,oBAAoB,EAEpB,WAAW,EACZ,MAAM,gBAAgB,CAAC;AAiBxB;;GAEG;AACH,wBAAgB,kBAAkB,CAChC,gBAAgB,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAChD,IAAI,EAAE,oBAAoB,GAAG,WAAW,CAAC,gBAAgB,CAAC,CA4B3D"}
@@ -1 +1 @@
1
- {"version":3,"file":"createClassifierFn.js","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":";;;;;;;;;;;;;AA2BA,gDA8BC;AAzDD,0CAA6C;AAQ7C,qEAAkE;AAElE;;;GAGG;AACH,SAAS,eAAe,CACtB,OAAiC;IAEjC,MAAM,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACzC,CAAC;IACD,OAAO,MAA+B,CAAC;AACzC,CAAC;AAED;;GAEG;AACH,SAAgB,kBAAkB,CAEhC,IAA0B;IAC1B,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,cAAc,KAAc,IAAI,EAAb,IAAI,UAAK,IAAI,EAAlD,sCAA2C,CAAO,CAAC;IAEzD,OAAO,KAAK,EAAE,IAAsB,EAA6B,EAAE;QACjE,MAAM,iBAAiB,qBAClB,IAAI,CACR,CAAC;QAEF,MAAM,MAAM,GAAG,IAAA,yBAAc,EAAC;YAC5B,QAAQ,EAAE,cAAc;YACxB,SAAS,EAAE,iBAAiB;SAC7B,CAAC,CAAC;QAEH,MAAM,cAAc,GAAG,MAAM,IAAA,+CAAsB,kBACjD,KAAK,EACL,MAAM,EAAE,eAAe,CAAC,OAAO,CAAC,EAChC,MAAM,IACH,IAAI,EACP,CAAC;QAEH,mEAAmE;QACnE,MAAM,KAAK,GAAG,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QAE5C,uBACE,KAAK,IACF,cAAc,EACjB;IACJ,CAAC,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"createClassifierFn.js","sourceRoot":"","sources":["../../../src/llm/createClassifierFn.ts"],"names":[],"mappings":";;;;;;;;;;;;;AA0BA,gDA8BC;AAxDD,0CAA6C;AAO7C,qEAAkE;AAElE;;;GAGG;AACH,SAAS,eAAe,CACtB,OAAiC;IAEjC,MAAM,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IACpC,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACzC,CAAC;IACD,OAAO,MAA+B,CAAC;AACzC,CAAC;AAED;;GAEG;AACH,SAAgB,kBAAkB,CAEhC,IAA0B;IAC1B,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,cAAc,KAAc,IAAI,EAAb,IAAI,UAAK,IAAI,EAAlD,sCAA2C,CAAO,CAAC;IAEzD,OAAO,KAAK,EAAE,IAAsB,EAA6B,EAAE;QACjE,MAAM,iBAAiB,qBAClB,IAAI,CACR,CAAC;QAEF,MAAM,MAAM,GAAG,IAAA,yBAAc,EAAC;YAC5B,QAAQ,EAAE,cAAc;YACxB,SAAS,EAAE,iBAAiB;SAC7B,CAAC,CAAC;QAEH,MAAM,cAAc,GAAG,MAAM,IAAA,+CAAsB,kBACjD,KAAK,EACL,MAAM,EAAE,eAAe,CAAC,OAAO,CAAC,EAChC,MAAM,IACH,IAAI,EACP,CAAC;QAEH,mEAAmE;QACnE,MAAM,KAAK,GAAG,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QAE5C,uBACE,KAAK,IACF,cAAc,EACjB;IACJ,CAAC,CAAC;AACJ,CAAC"}
@@ -0,0 +1,43 @@
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
3
+ export interface ConcisenessEvaluatorArgs<RecordType extends Record<string, unknown> = ConcisenessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
4
+ optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
5
+ name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
6
+ choices?: CreateClassificationEvaluatorArgs<RecordType>["choices"];
7
+ promptTemplate?: CreateClassificationEvaluatorArgs<RecordType>["promptTemplate"];
8
+ }
9
+ /**
10
+ * A record to be evaluated by the conciseness evaluator.
11
+ */
12
+ export type ConcisenessEvaluationRecord = {
13
+ input: string;
14
+ output: string;
15
+ };
16
+ /**
17
+ * Creates a conciseness evaluator function.
18
+ *
19
+ * This function returns an evaluator that determines whether a given output
20
+ * is concise and free of unnecessary content such as pleasantries, hedging,
21
+ * meta-commentary, or redundant information.
22
+ *
23
+ * @param args - The arguments for creating the conciseness evaluator.
24
+ * @param args.model - The model to use for classification.
25
+ * @param args.choices - The possible classification choices (defaults to CONCISENESS_CHOICES).
26
+ * @param args.promptTemplate - The prompt template to use (defaults to CONCISENESS_TEMPLATE).
27
+ * @param args.telemetry - The telemetry to use for the evaluator.
28
+ *
29
+ * @returns An evaluator function that takes a {@link ConcisenessEvaluationRecord} and returns a classification result
30
+ * indicating whether the output is concise or verbose.
31
+ *
32
+ * @example
33
+ * ```ts
34
+ * const evaluator = createConcisenessEvaluator({ model: openai("gpt-4o-mini") });
35
+ * const result = await evaluator.evaluate({
36
+ * input: "What is the capital of France?",
37
+ * output: "Paris.",
38
+ * });
39
+ * console.log(result.label); // "concise" or "verbose"
40
+ * ```
41
+ */
42
+ export declare function createConcisenessEvaluator<RecordType extends Record<string, unknown> = ConcisenessEvaluationRecord>(args: ConcisenessEvaluatorArgs<RecordType>): ClassificationEvaluator<RecordType>;
43
+ //# sourceMappingURL=createConcisenessEvaluator.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"createConcisenessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createConcisenessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,wBAAwB,CACvC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,CACxE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,0BAA0B,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,EAExE,IAAI,EAAE,wBAAwB,CAAC,UAAU,CAAC,GACzC,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
@@ -0,0 +1,50 @@
1
+ "use strict";
2
+ var __rest = (this && this.__rest) || function (s, e) {
3
+ var t = {};
4
+ for (var p in s) if (Object.prototype.hasOwnProperty.call(s, p) && e.indexOf(p) < 0)
5
+ t[p] = s[p];
6
+ if (s != null && typeof Object.getOwnPropertySymbols === "function")
7
+ for (var i = 0, p = Object.getOwnPropertySymbols(s); i < p.length; i++) {
8
+ if (e.indexOf(p[i]) < 0 && Object.prototype.propertyIsEnumerable.call(s, p[i]))
9
+ t[p[i]] = s[p[i]];
10
+ }
11
+ return t;
12
+ };
13
+ Object.defineProperty(exports, "__esModule", { value: true });
14
+ exports.createConcisenessEvaluator = createConcisenessEvaluator;
15
+ const default_templates_1 = require("../__generated__/default_templates");
16
+ const createClassificationEvaluator_1 = require("./createClassificationEvaluator");
17
+ /**
18
+ * Creates a conciseness evaluator function.
19
+ *
20
+ * This function returns an evaluator that determines whether a given output
21
+ * is concise and free of unnecessary content such as pleasantries, hedging,
22
+ * meta-commentary, or redundant information.
23
+ *
24
+ * @param args - The arguments for creating the conciseness evaluator.
25
+ * @param args.model - The model to use for classification.
26
+ * @param args.choices - The possible classification choices (defaults to CONCISENESS_CHOICES).
27
+ * @param args.promptTemplate - The prompt template to use (defaults to CONCISENESS_TEMPLATE).
28
+ * @param args.telemetry - The telemetry to use for the evaluator.
29
+ *
30
+ * @returns An evaluator function that takes a {@link ConcisenessEvaluationRecord} and returns a classification result
31
+ * indicating whether the output is concise or verbose.
32
+ *
33
+ * @example
34
+ * ```ts
35
+ * const evaluator = createConcisenessEvaluator({ model: openai("gpt-4o-mini") });
36
+ * const result = await evaluator.evaluate({
37
+ * input: "What is the capital of France?",
38
+ * output: "Paris.",
39
+ * });
40
+ * console.log(result.label); // "concise" or "verbose"
41
+ * ```
42
+ */
43
+ function createConcisenessEvaluator(args) {
44
+ const { choices = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.choices, promptTemplate = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.template, optimizationDirection = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection, name = default_templates_1.CONCISENESS_CLASSIFICATION_EVALUATOR_CONFIG.name } = args, rest = __rest(args, ["choices", "promptTemplate", "optimizationDirection", "name"]);
45
+ return (0, createClassificationEvaluator_1.createClassificationEvaluator)(Object.assign(Object.assign({}, rest), { promptTemplate,
46
+ choices,
47
+ optimizationDirection,
48
+ name }));
49
+ }
50
+ //# sourceMappingURL=createConcisenessEvaluator.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"createConcisenessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createConcisenessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAmDA,gEAmBC;AAtED,0EAAiG;AAGjG,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAgB,0BAA0B,CAGxC,IAA0C;IAE1C,MAAM,EACJ,OAAO,GAAG,+DAA2C,CAAC,OAAO,EAC7D,cAAc,GAAG,+DAA2C,CAAC,QAAQ,EACrE,qBAAqB,GAAG,+DAA2C,CAAC,qBAAqB,EACzF,IAAI,GAAG,+DAA2C,CAAC,IAAI,KAErD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
@@ -1,5 +1,5 @@
1
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
2
- import { ClassificationEvaluator } from "./ClassificationEvaluator";
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
3
3
  export interface CorrectnessEvaluatorArgs<RecordType extends Record<string, unknown> = CorrectnessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
4
4
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
5
5
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
@@ -1 +1 @@
1
- {"version":3,"file":"createCorrectnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGpE,MAAM,WAAW,wBAAwB,CACvC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,CACxE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,wBAAgB,0BAA0B,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,EAExE,IAAI,EAAE,wBAAwB,CAAC,UAAU,CAAC,GACzC,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
1
+ {"version":3,"file":"createCorrectnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,wBAAwB,CACvC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,CACxE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,wBAAgB,0BAA0B,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,2BAA2B,EAExE,IAAI,EAAE,wBAAwB,CAAC,UAAU,CAAC,GACzC,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createCorrectnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAmDA,gEAmBC;AAtED,0EAAiG;AAIjG,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,SAAgB,0BAA0B,CAGxC,IAA0C;IAE1C,MAAM,EACJ,OAAO,GAAG,+DAA2C,CAAC,OAAO,EAC7D,cAAc,GAAG,+DAA2C,CAAC,QAAQ,EACrE,qBAAqB,GAAG,+DAA2C,CAAC,qBAAqB,EACzF,IAAI,GAAG,+DAA2C,CAAC,IAAI,KAErD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createCorrectnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createCorrectnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAkDA,gEAmBC;AArED,0EAAiG;AAGjG,mFAAgF;AAsBhF;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,SAAgB,0BAA0B,CAGxC,IAA0C;IAE1C,MAAM,EACJ,OAAO,GAAG,+DAA2C,CAAC,OAAO,EAC7D,cAAc,GAAG,+DAA2C,CAAC,QAAQ,EACrE,qBAAqB,GAAG,+DAA2C,CAAC,qBAAqB,EACzF,IAAI,GAAG,+DAA2C,CAAC,IAAI,KAErD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
@@ -1,5 +1,5 @@
1
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
2
- import { ClassificationEvaluator } from "./ClassificationEvaluator";
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
3
3
  export interface DocumentRelevanceEvaluatorArgs<RecordType extends Record<string, unknown> = DocumentRelevanceEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
4
4
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
5
5
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
@@ -1 +1 @@
1
- {"version":3,"file":"createDocumentRelevanceEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGpE,MAAM,WAAW,8BAA8B,CAC7C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,CACnC,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,WAAW,iCAAiC;IAChD,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,gCAAgC,CAC9C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,EAEnC,IAAI,EAAE,8BAA8B,CAAC,UAAU,CAAC,GAC/C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
1
+ {"version":3,"file":"createDocumentRelevanceEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,8BAA8B,CAC7C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,CACnC,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,WAAW,iCAAiC;IAChD,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,wBAAgB,gCAAgC,CAC9C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACxC,iCAAiC,EAEnC,IAAI,EAAE,8BAA8B,CAAC,UAAU,CAAC,GAC/C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createDocumentRelevanceEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAsDA,4EAoBC;AA1ED,0EAAwG;AAIxG,mFAAgF;AAwBhF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAgB,gCAAgC,CAI9C,IAAgD;IAEhD,MAAM,EACJ,OAAO,GAAG,sEAAkD,CAAC,OAAO,EACpE,cAAc,GAAG,sEAAkD,CAAC,QAAQ,EAC5E,qBAAqB,GAAG,sEAAkD,CAAC,qBAAqB,EAChG,IAAI,GAAG,sEAAkD,CAAC,IAAI,KAE5D,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createDocumentRelevanceEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createDocumentRelevanceEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAqDA,4EAoBC;AAzED,0EAAwG;AAGxG,mFAAgF;AAwBhF;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAgB,gCAAgC,CAI9C,IAAgD;IAEhD,MAAM,EACJ,OAAO,GAAG,sEAAkD,CAAC,OAAO,EACpE,cAAc,GAAG,sEAAkD,CAAC,QAAQ,EAC5E,qBAAqB,GAAG,sEAAkD,CAAC,qBAAqB,EAChG,IAAI,GAAG,sEAAkD,CAAC,IAAI,KAE5D,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
@@ -1,5 +1,5 @@
1
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
2
- import { ClassificationEvaluator } from "./ClassificationEvaluator";
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
3
3
  export interface FaithfulnessEvaluatorArgs<RecordType extends Record<string, unknown> = FaithfulnessEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
4
4
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
5
5
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
@@ -1 +1 @@
1
- {"version":3,"file":"createFaithfulnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGpE,MAAM,WAAW,yBAAyB,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,CACzE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,4BAA4B,GAAG;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AACF;;;;;GAKG;AACH,wBAAgB,2BAA2B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,EAEzE,IAAI,EAAE,yBAAyB,CAAC,UAAU,CAAC,GAC1C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
1
+ {"version":3,"file":"createFaithfulnessEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,yBAAyB,CACxC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,CACzE,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,4BAA4B,GAAG;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AACF;;;;;GAKG;AACH,wBAAgB,2BAA2B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,4BAA4B,EAEzE,IAAI,EAAE,yBAAyB,CAAC,UAAU,CAAC,GAC1C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createFaithfulnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAgCA,kEAmBC;AAnDD,0EAAkG;AAIlG,mFAAgF;AAsBhF;;;;;GAKG;AACH,SAAgB,2BAA2B,CAGzC,IAA2C;IAE3C,MAAM,EACJ,OAAO,GAAG,gEAA4C,CAAC,OAAO,EAC9D,cAAc,GAAG,gEAA4C,CAAC,QAAQ,EACtE,qBAAqB,GAAG,gEAA4C,CAAC,qBAAqB,EAC1F,IAAI,GAAG,gEAA4C,CAAC,IAAI,KAEtD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createFaithfulnessEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createFaithfulnessEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AA+BA,kEAmBC;AAlDD,0EAAkG;AAGlG,mFAAgF;AAsBhF;;;;;GAKG;AACH,SAAgB,2BAA2B,CAGzC,IAA2C;IAE3C,MAAM,EACJ,OAAO,GAAG,gEAA4C,CAAC,OAAO,EAC9D,cAAc,GAAG,gEAA4C,CAAC,QAAQ,EACtE,qBAAqB,GAAG,gEAA4C,CAAC,qBAAqB,EAC1F,IAAI,GAAG,gEAA4C,CAAC,IAAI,KAEtD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
@@ -4,8 +4,8 @@
4
4
  * - 'faithful'/'unfaithful' labels instead of 'factual'/'hallucinated'
5
5
  * - Maximizes score (1.0=faithful) instead of minimizing it
6
6
  */
7
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
8
- import { ClassificationEvaluator } from "./ClassificationEvaluator";
7
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
8
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
9
9
  export interface HallucinationEvaluatorArgs<RecordType extends Record<string, unknown> = HallucinationEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
10
10
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
11
11
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
@@ -1 +1 @@
1
- {"version":3,"file":"createHallucinationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGpE,MAAM,WAAW,0BAA0B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,CAC1E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,6BAA6B,GAAG;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAgB,4BAA4B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,EAE1E,IAAI,EAAE,0BAA0B,CAAC,UAAU,CAAC,GAC3C,uBAAuB,CAAC,UAAU,CAAC,CAuBrC"}
1
+ {"version":3,"file":"createHallucinationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,0BAA0B,CACzC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,CAC1E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,6BAA6B,GAAG;IAC1C,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAgB,4BAA4B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,6BAA6B,EAE1E,IAAI,EAAE,0BAA0B,CAAC,UAAU,CAAC,GAC3C,uBAAuB,CAAC,UAAU,CAAC,CAuBrC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createHallucinationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;;;;;;;;;;;AAyCH,oEA2BC;AAlED,0EAAmG;AAInG,mFAAgF;AAuBhF;;;;;;;;;;;GAWG;AACH,SAAgB,4BAA4B,CAG1C,IAA4C;IAE5C,sCAAsC;IACtC,OAAO,CAAC,IAAI,CACV,sFAAsF;QACpF,yEAAyE;QACzE,+EAA+E;QAC/E,8BAA8B,CACjC,CAAC;IAEF,MAAM,EACJ,OAAO,GAAG,iEAA6C,CAAC,OAAO,EAC/D,cAAc,GAAG,iEAA6C,CAAC,QAAQ,EACvE,qBAAqB,GAAG,iEAA6C,CAAC,qBAAqB,EAC3F,IAAI,GAAG,iEAA6C,CAAC,IAAI,KAEvD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createHallucinationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createHallucinationEvaluator.ts"],"names":[],"mappings":";AAAA;;;;;GAKG;;;;;;;;;;;;;AAwCH,oEA2BC;AAjED,0EAAmG;AAGnG,mFAAgF;AAuBhF;;;;;;;;;;;GAWG;AACH,SAAgB,4BAA4B,CAG1C,IAA4C;IAE5C,sCAAsC;IACtC,OAAO,CAAC,IAAI,CACV,sFAAsF;QACpF,yEAAyE;QACzE,+EAA+E;QAC/E,8BAA8B,CACjC,CAAC;IAEF,MAAM,EACJ,OAAO,GAAG,iEAA6C,CAAC,OAAO,EAC/D,cAAc,GAAG,iEAA6C,CAAC,QAAQ,EACvE,qBAAqB,GAAG,iEAA6C,CAAC,qBAAqB,EAC3F,IAAI,GAAG,iEAA6C,CAAC,IAAI,KAEvD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
@@ -1,5 +1,5 @@
1
- import { CreateClassificationEvaluatorArgs } from "../types/evals";
2
- import { ClassificationEvaluator } from "./ClassificationEvaluator";
1
+ import type { CreateClassificationEvaluatorArgs } from "../types/evals";
2
+ import type { ClassificationEvaluator } from "./ClassificationEvaluator";
3
3
  export interface ToolInvocationEvaluatorArgs<RecordType extends Record<string, unknown> = ToolInvocationEvaluationRecord> extends Omit<CreateClassificationEvaluatorArgs<RecordType>, "promptTemplate" | "choices" | "optimizationDirection" | "name"> {
4
4
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
5
5
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
@@ -1 +1 @@
1
- {"version":3,"file":"createToolInvocationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AAEnE,OAAO,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGpE,MAAM,WAAW,2BAA2B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,CAC3E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,8BAA8B,GAAG;IAC3C;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;OAEG;IACH,aAAa,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AACH,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,EAE3E,IAAI,EAAE,2BAA2B,CAAC,UAAU,CAAC,GAC5C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
1
+ {"version":3,"file":"createToolInvocationEvaluator.d.ts","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,2BAA2B,CAAC;AAGzE,MAAM,WAAW,2BAA2B,CAC1C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,CAC3E,SAAQ,IAAI,CACZ,iCAAiC,CAAC,UAAU,CAAC,EAC7C,gBAAgB,GAAG,SAAS,GAAG,uBAAuB,GAAG,MAAM,CAChE;IACC,qBAAqB,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,uBAAuB,CAAC,CAAC;IAC/F,IAAI,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;IAC7D,OAAO,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC;IACnE,cAAc,CAAC,EAAE,iCAAiC,CAAC,UAAU,CAAC,CAAC,gBAAgB,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,8BAA8B,GAAG;IAC3C;;OAEG;IACH,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,cAAc,EAAE,MAAM,CAAC;IACvB;;OAEG;IACH,aAAa,EAAE,MAAM,CAAC;CACvB,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AACH,wBAAgB,6BAA6B,CAC3C,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,8BAA8B,EAE3E,IAAI,EAAE,2BAA2B,CAAC,UAAU,CAAC,GAC5C,uBAAuB,CAAC,UAAU,CAAC,CAerC"}
@@ -1 +1 @@
1
- {"version":3,"file":"createToolInvocationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAmFA,sEAmBC;AAtGD,0EAAqG;AAIrG,mFAAgF;AAgChF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AACH,SAAgB,6BAA6B,CAG3C,IAA6C;IAE7C,MAAM,EACJ,OAAO,GAAG,mEAA+C,CAAC,OAAO,EACjE,cAAc,GAAG,mEAA+C,CAAC,QAAQ,EACzE,qBAAqB,GAAG,mEAA+C,CAAC,qBAAqB,EAC7F,IAAI,GAAG,mEAA+C,CAAC,IAAI,KAEzD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}
1
+ {"version":3,"file":"createToolInvocationEvaluator.js","sourceRoot":"","sources":["../../../src/llm/createToolInvocationEvaluator.ts"],"names":[],"mappings":";;;;;;;;;;;;;AAkFA,sEAmBC;AArGD,0EAAqG;AAGrG,mFAAgF;AAgChF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8CG;AACH,SAAgB,6BAA6B,CAG3C,IAA6C;IAE7C,MAAM,EACJ,OAAO,GAAG,mEAA+C,CAAC,OAAO,EACjE,cAAc,GAAG,mEAA+C,CAAC,QAAQ,EACzE,qBAAqB,GAAG,mEAA+C,CAAC,qBAAqB,EAC7F,IAAI,GAAG,mEAA+C,CAAC,IAAI,KAEzD,IAAI,EADH,IAAI,UACL,IAAI,EANF,8DAML,CAAO,CAAC;IACT,OAAO,IAAA,6DAA6B,kCAC/B,IAAI,KACP,cAAc;QACd,OAAO;QACP,qBAAqB;QACrB,IAAI,IACJ,CAAC;AACL,CAAC"}