promptfoo 0.66.0 → 0.67.0

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 (281) hide show
  1. package/README.md +9 -9
  2. package/dist/package.json +3 -2
  3. package/dist/src/assertions.d.ts.map +1 -1
  4. package/dist/src/assertions.js +18 -47
  5. package/dist/src/assertions.js.map +1 -1
  6. package/dist/src/cache.d.ts.map +1 -1
  7. package/dist/src/cache.js +3 -3
  8. package/dist/src/cache.js.map +1 -1
  9. package/dist/src/commands/config.d.ts.map +1 -1
  10. package/dist/src/commands/config.js +13 -2
  11. package/dist/src/commands/config.js.map +1 -1
  12. package/dist/src/commands/delete.js +1 -1
  13. package/dist/src/commands/delete.js.map +1 -1
  14. package/dist/src/commands/export.d.ts.map +1 -1
  15. package/dist/src/commands/export.js +1 -1
  16. package/dist/src/commands/export.js.map +1 -1
  17. package/dist/src/commands/import.js +1 -1
  18. package/dist/src/commands/import.js.map +1 -1
  19. package/dist/src/commands/list.d.ts.map +1 -1
  20. package/dist/src/commands/list.js +2 -2
  21. package/dist/src/commands/list.js.map +1 -1
  22. package/dist/src/commands/show.d.ts.map +1 -1
  23. package/dist/src/commands/show.js +2 -2
  24. package/dist/src/commands/show.js.map +1 -1
  25. package/dist/src/csv.d.ts.map +1 -1
  26. package/dist/src/csv.js +0 -1
  27. package/dist/src/csv.js.map +1 -1
  28. package/dist/src/database.d.ts.map +1 -1
  29. package/dist/src/database.js +5 -5
  30. package/dist/src/database.js.map +1 -1
  31. package/dist/src/esm.d.ts.map +1 -1
  32. package/dist/src/esm.js +1 -1
  33. package/dist/src/esm.js.map +1 -1
  34. package/dist/src/evaluator.d.ts +18 -0
  35. package/dist/src/evaluator.d.ts.map +1 -1
  36. package/dist/src/evaluator.js +41 -23
  37. package/dist/src/evaluator.js.map +1 -1
  38. package/dist/src/feedback.d.ts.map +1 -1
  39. package/dist/src/feedback.js +2 -2
  40. package/dist/src/feedback.js.map +1 -1
  41. package/dist/src/fetch.d.ts.map +1 -1
  42. package/dist/src/fetch.js.map +1 -1
  43. package/dist/src/globalConfig.d.ts +0 -4
  44. package/dist/src/globalConfig.d.ts.map +1 -1
  45. package/dist/src/globalConfig.js +5 -5
  46. package/dist/src/globalConfig.js.map +1 -1
  47. package/dist/src/googleSheets.d.ts.map +1 -1
  48. package/dist/src/googleSheets.js +6 -6
  49. package/dist/src/googleSheets.js.map +1 -1
  50. package/dist/src/index.d.ts +1 -1
  51. package/dist/src/index.d.ts.map +1 -1
  52. package/dist/src/index.js +4 -4
  53. package/dist/src/index.js.map +1 -1
  54. package/dist/src/integrations/langfuse.js +1 -1
  55. package/dist/src/integrations/langfuse.js.map +1 -1
  56. package/dist/src/main.js +26 -26
  57. package/dist/src/main.js.map +1 -1
  58. package/dist/src/matchers.d.ts.map +1 -1
  59. package/dist/src/matchers.js +8 -7
  60. package/dist/src/matchers.js.map +1 -1
  61. package/dist/src/migrate.d.ts.map +1 -1
  62. package/dist/src/migrate.js +2 -2
  63. package/dist/src/migrate.js.map +1 -1
  64. package/dist/src/onboarding.d.ts.map +1 -1
  65. package/dist/src/onboarding.js +2 -2
  66. package/dist/src/onboarding.js.map +1 -1
  67. package/dist/src/prompts/constants.d.ts +3 -0
  68. package/dist/src/prompts/constants.d.ts.map +1 -0
  69. package/dist/src/prompts/constants.js +16 -0
  70. package/dist/src/prompts/constants.js.map +1 -0
  71. package/dist/src/prompts/external/ragas.d.ts.map +1 -0
  72. package/dist/src/prompts/external/ragas.js.map +1 -0
  73. package/dist/src/{prompts.d.ts → prompts/grading.d.ts} +7 -4
  74. package/dist/src/prompts/grading.d.ts.map +1 -0
  75. package/dist/src/prompts/grading.js +119 -0
  76. package/dist/src/prompts/grading.js.map +1 -0
  77. package/dist/src/prompts/index.d.ts +25 -0
  78. package/dist/src/prompts/index.d.ts.map +1 -0
  79. package/dist/src/prompts/index.js +143 -0
  80. package/dist/src/prompts/index.js.map +1 -0
  81. package/dist/src/prompts/processors/javascript.d.ts +9 -0
  82. package/dist/src/prompts/processors/javascript.d.ts.map +1 -0
  83. package/dist/src/prompts/processors/javascript.js +22 -0
  84. package/dist/src/prompts/processors/javascript.js.map +1 -0
  85. package/dist/src/prompts/processors/json.d.ts +12 -0
  86. package/dist/src/prompts/processors/json.d.ts.map +1 -0
  87. package/dist/src/prompts/processors/json.js +48 -0
  88. package/dist/src/prompts/processors/json.js.map +1 -0
  89. package/dist/src/prompts/processors/jsonl.d.ts +9 -0
  90. package/dist/src/prompts/processors/jsonl.d.ts.map +1 -0
  91. package/dist/src/prompts/processors/jsonl.js +48 -0
  92. package/dist/src/prompts/processors/jsonl.js.map +1 -0
  93. package/dist/src/prompts/processors/python.d.ts +31 -0
  94. package/dist/src/prompts/processors/python.d.ts.map +1 -0
  95. package/dist/src/prompts/processors/python.js +96 -0
  96. package/dist/src/prompts/processors/python.js.map +1 -0
  97. package/dist/src/prompts/processors/string.d.ts +8 -0
  98. package/dist/src/prompts/processors/string.d.ts.map +1 -0
  99. package/dist/src/prompts/processors/string.js +23 -0
  100. package/dist/src/prompts/processors/string.js.map +1 -0
  101. package/dist/src/prompts/processors/text.d.ts +9 -0
  102. package/dist/src/prompts/processors/text.d.ts.map +1 -0
  103. package/dist/src/prompts/processors/text.js +47 -0
  104. package/dist/src/prompts/processors/text.js.map +1 -0
  105. package/dist/src/prompts/processors/yaml.d.ts +13 -0
  106. package/dist/src/prompts/processors/yaml.d.ts.map +1 -0
  107. package/dist/src/prompts/processors/yaml.js +55 -0
  108. package/dist/src/prompts/processors/yaml.js.map +1 -0
  109. package/dist/src/prompts/utils.d.ts +29 -0
  110. package/dist/src/prompts/utils.d.ts.map +1 -0
  111. package/dist/src/prompts/utils.js +143 -0
  112. package/dist/src/prompts/utils.js.map +1 -0
  113. package/dist/src/providers/anthropic.d.ts.map +1 -1
  114. package/dist/src/providers/anthropic.js +1 -1
  115. package/dist/src/providers/anthropic.js.map +1 -1
  116. package/dist/src/providers/azureopenai.d.ts.map +1 -1
  117. package/dist/src/providers/azureopenai.js +1 -1
  118. package/dist/src/providers/azureopenai.js.map +1 -1
  119. package/dist/src/providers/azureopenaiUtil.d.ts.map +1 -1
  120. package/dist/src/providers/azureopenaiUtil.js +2 -2
  121. package/dist/src/providers/azureopenaiUtil.js.map +1 -1
  122. package/dist/src/providers/bam.d.ts.map +1 -1
  123. package/dist/src/providers/bam.js +1 -1
  124. package/dist/src/providers/bam.js.map +1 -1
  125. package/dist/src/providers/bedrock.d.ts +1 -0
  126. package/dist/src/providers/bedrock.d.ts.map +1 -1
  127. package/dist/src/providers/bedrock.js +21 -4
  128. package/dist/src/providers/bedrock.js.map +1 -1
  129. package/dist/src/providers/cloudflare-ai.d.ts +1 -1
  130. package/dist/src/providers/cloudflare-ai.d.ts.map +1 -1
  131. package/dist/src/providers/cloudflare-ai.js +1 -1
  132. package/dist/src/providers/cloudflare-ai.js.map +1 -1
  133. package/dist/src/providers/cohere.d.ts.map +1 -1
  134. package/dist/src/providers/cohere.js.map +1 -1
  135. package/dist/src/providers/defaults.d.ts +1 -1
  136. package/dist/src/providers/defaults.d.ts.map +1 -1
  137. package/dist/src/providers/defaults.js +2 -2
  138. package/dist/src/providers/defaults.js.map +1 -1
  139. package/dist/src/providers/http.d.ts.map +1 -1
  140. package/dist/src/providers/http.js +2 -2
  141. package/dist/src/providers/http.js.map +1 -1
  142. package/dist/src/providers/huggingface.d.ts +1 -1
  143. package/dist/src/providers/huggingface.d.ts.map +1 -1
  144. package/dist/src/providers/huggingface.js +1 -1
  145. package/dist/src/providers/huggingface.js.map +1 -1
  146. package/dist/src/providers/llama.d.ts.map +1 -1
  147. package/dist/src/providers/llama.js.map +1 -1
  148. package/dist/src/providers/localai.d.ts.map +1 -1
  149. package/dist/src/providers/localai.js +1 -1
  150. package/dist/src/providers/localai.js.map +1 -1
  151. package/dist/src/providers/mistral.d.ts.map +1 -1
  152. package/dist/src/providers/mistral.js +1 -1
  153. package/dist/src/providers/mistral.js.map +1 -1
  154. package/dist/src/providers/ollama.d.ts.map +1 -1
  155. package/dist/src/providers/ollama.js +1 -1
  156. package/dist/src/providers/ollama.js.map +1 -1
  157. package/dist/src/providers/openai.d.ts +1 -1
  158. package/dist/src/providers/openai.d.ts.map +1 -1
  159. package/dist/src/providers/openai.js +2 -2
  160. package/dist/src/providers/openai.js.map +1 -1
  161. package/dist/src/providers/palm.d.ts.map +1 -1
  162. package/dist/src/providers/palm.js +1 -1
  163. package/dist/src/providers/palm.js.map +1 -1
  164. package/dist/src/providers/portkey.d.ts +1 -1
  165. package/dist/src/providers/portkey.d.ts.map +1 -1
  166. package/dist/src/providers/portkey.js.map +1 -1
  167. package/dist/src/providers/promptfoo.d.ts.map +1 -1
  168. package/dist/src/providers/promptfoo.js.map +1 -1
  169. package/dist/src/providers/pythonCompletion.d.ts.map +1 -1
  170. package/dist/src/providers/pythonCompletion.js +2 -2
  171. package/dist/src/providers/pythonCompletion.js.map +1 -1
  172. package/dist/src/providers/replicate.d.ts.map +1 -1
  173. package/dist/src/providers/replicate.js +2 -2
  174. package/dist/src/providers/replicate.js.map +1 -1
  175. package/dist/src/providers/scriptCompletion.d.ts.map +1 -1
  176. package/dist/src/providers/scriptCompletion.js +1 -1
  177. package/dist/src/providers/scriptCompletion.js.map +1 -1
  178. package/dist/src/providers/vertex.d.ts +12 -1
  179. package/dist/src/providers/vertex.d.ts.map +1 -1
  180. package/dist/src/providers/vertex.js +60 -3
  181. package/dist/src/providers/vertex.js.map +1 -1
  182. package/dist/src/providers/voyage.d.ts.map +1 -1
  183. package/dist/src/providers/voyage.js.map +1 -1
  184. package/dist/src/providers/webhook.d.ts.map +1 -1
  185. package/dist/src/providers/webhook.js +1 -1
  186. package/dist/src/providers/webhook.js.map +1 -1
  187. package/dist/src/providers.d.ts +3 -3
  188. package/dist/src/providers.d.ts.map +1 -1
  189. package/dist/src/providers.js +30 -20
  190. package/dist/src/providers.js.map +1 -1
  191. package/dist/src/python/wrapper.d.ts.map +1 -1
  192. package/dist/src/python/wrapper.js +1 -1
  193. package/dist/src/python/wrapper.js.map +1 -1
  194. package/dist/src/redteam/getHijackingTests.d.ts.map +1 -1
  195. package/dist/src/redteam/getHijackingTests.js.map +1 -1
  196. package/dist/src/redteam/index.d.ts.map +1 -1
  197. package/dist/src/redteam/index.js +4 -4
  198. package/dist/src/redteam/index.js.map +1 -1
  199. package/dist/src/redteam/iterative.d.ts +3 -0
  200. package/dist/src/redteam/iterative.d.ts.map +1 -1
  201. package/dist/src/redteam/iterative.js +10 -3
  202. package/dist/src/redteam/iterative.js.map +1 -1
  203. package/dist/src/share.d.ts.map +1 -1
  204. package/dist/src/share.js +1 -1
  205. package/dist/src/share.js.map +1 -1
  206. package/dist/src/suggestions.d.ts.map +1 -1
  207. package/dist/src/suggestions.js.map +1 -1
  208. package/dist/src/table.d.ts.map +1 -1
  209. package/dist/src/table.js +1 -1
  210. package/dist/src/table.js.map +1 -1
  211. package/dist/src/telemetry.d.ts.map +1 -1
  212. package/dist/src/telemetry.js +1 -1
  213. package/dist/src/telemetry.js.map +1 -1
  214. package/dist/src/testCases.d.ts.map +1 -1
  215. package/dist/src/testCases.js +5 -5
  216. package/dist/src/testCases.js.map +1 -1
  217. package/dist/src/types.d.ts +4 -1
  218. package/dist/src/types.d.ts.map +1 -1
  219. package/dist/src/types.js.map +1 -1
  220. package/dist/src/updates.d.ts.map +1 -1
  221. package/dist/src/updates.js +2 -2
  222. package/dist/src/updates.js.map +1 -1
  223. package/dist/src/util.d.ts +1 -0
  224. package/dist/src/util.d.ts.map +1 -1
  225. package/dist/src/util.js +44 -14
  226. package/dist/src/util.js.map +1 -1
  227. package/dist/src/web/nextui/404/index.html +1 -1
  228. package/dist/src/web/nextui/404.html +1 -1
  229. package/dist/src/web/nextui/_next/static/chunks/2-f562fcee36045aca.js +1 -0
  230. package/dist/src/web/nextui/_next/static/chunks/app/auth/login/{page-c4a2650ac3a0ecd9.js → page-d932a73274f0f175.js} +1 -1
  231. package/dist/src/web/nextui/_next/static/chunks/app/auth/signup/{page-dd18caf3100d8d0e.js → page-7a8f35189f8bc5b8.js} +1 -1
  232. package/dist/src/web/nextui/_next/static/chunks/app/datasets/page-9712a72be5b92b81.js +1 -0
  233. package/dist/src/web/nextui/_next/static/chunks/app/layout-c6a855ac7037db70.js +1 -0
  234. package/dist/src/web/nextui/_next/static/chunks/app/progress/{page-00b5c3308a81af12.js → page-751105ef02d00993.js} +1 -1
  235. package/dist/src/web/nextui/_next/static/chunks/app/prompts/page-689202b79a6b6a92.js +1 -0
  236. package/dist/src/web/nextui/_next/static/chunks/app/report/page-4b1c00ce9ee1f5ab.js +1 -0
  237. package/dist/src/web/nextui/_next/static/chunks/app/setup/page-544af5b41c35d73a.js +1 -0
  238. package/dist/src/web/nextui/_next/static/chunks/webpack-d1a82d85a184970b.js +1 -0
  239. package/dist/src/web/nextui/_next/static/css/{51a17e8edcdfdbb2.css → b76f1739b5a14901.css} +1 -1
  240. package/dist/src/web/nextui/_next/static/css/dab5d695b3657d59.css +1 -0
  241. package/dist/src/web/nextui/_next/static/css/edcd6f0b6c902fde.css +1 -0
  242. package/dist/src/web/nextui/auth/login/index.html +1 -1
  243. package/dist/src/web/nextui/auth/login/index.txt +6 -6
  244. package/dist/src/web/nextui/auth/signup/index.html +1 -1
  245. package/dist/src/web/nextui/auth/signup/index.txt +6 -6
  246. package/dist/src/web/nextui/datasets/index.html +1 -1
  247. package/dist/src/web/nextui/datasets/index.txt +6 -6
  248. package/dist/src/web/nextui/eval/index.html +1 -1
  249. package/dist/src/web/nextui/eval/index.txt +6 -6
  250. package/dist/src/web/nextui/index.html +1 -1
  251. package/dist/src/web/nextui/index.txt +5 -5
  252. package/dist/src/web/nextui/progress/index.html +1 -1
  253. package/dist/src/web/nextui/progress/index.txt +6 -6
  254. package/dist/src/web/nextui/prompts/index.html +1 -1
  255. package/dist/src/web/nextui/prompts/index.txt +6 -6
  256. package/dist/src/web/nextui/report/index.html +1 -1
  257. package/dist/src/web/nextui/report/index.txt +8 -8
  258. package/dist/src/web/nextui/setup/index.html +2 -2
  259. package/dist/src/web/nextui/setup/index.txt +9 -9
  260. package/dist/src/web/server.d.ts.map +1 -1
  261. package/dist/src/web/server.js +10 -10
  262. package/dist/src/web/server.js.map +1 -1
  263. package/package.json +3 -2
  264. package/dist/src/external/ragas.d.ts.map +0 -1
  265. package/dist/src/external/ragas.js.map +0 -1
  266. package/dist/src/prompts.d.ts.map +0 -1
  267. package/dist/src/prompts.js +0 -391
  268. package/dist/src/prompts.js.map +0 -1
  269. package/dist/src/web/nextui/_next/static/chunks/2-60ab1c881a240da6.js +0 -1
  270. package/dist/src/web/nextui/_next/static/chunks/app/datasets/page-9426b519d4be1fdb.js +0 -1
  271. package/dist/src/web/nextui/_next/static/chunks/app/layout-dfda5ed5ef745c2d.js +0 -1
  272. package/dist/src/web/nextui/_next/static/chunks/app/prompts/page-ee610cffca4b965b.js +0 -1
  273. package/dist/src/web/nextui/_next/static/chunks/app/report/page-1b97ddc1b365a121.js +0 -1
  274. package/dist/src/web/nextui/_next/static/chunks/app/setup/page-e1c49ea6fe7c04c5.js +0 -1
  275. package/dist/src/web/nextui/_next/static/chunks/webpack-2fa22c6070dd15bc.js +0 -1
  276. package/dist/src/web/nextui/_next/static/css/16c1dd82fc87c9d7.css +0 -1
  277. package/dist/src/web/nextui/_next/static/css/51f7d6933894a4f8.css +0 -1
  278. /package/dist/src/{external → prompts/external}/ragas.d.ts +0 -0
  279. /package/dist/src/{external → prompts/external}/ragas.js +0 -0
  280. /package/dist/src/web/nextui/_next/static/{lMO8mRWL6KkcjtN4Giq14 → m-VSM7eKSuIZDyA_tnwFX}/_buildManifest.js +0 -0
  281. /package/dist/src/web/nextui/_next/static/{lMO8mRWL6KkcjtN4Giq14 → m-VSM7eKSuIZDyA_tnwFX}/_ssgManifest.js +0 -0
@@ -1 +1 @@
1
- <!DOCTYPE html><html><head><meta charSet="utf-8"/><meta name="viewport" content="width=device-width"/><title>404: This page could not be found</title><meta name="next-head-count" content="3"/><noscript data-n-css=""></noscript><script defer="" nomodule="" src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js"></script><script src="/_next/static/chunks/webpack-2fa22c6070dd15bc.js" defer=""></script><script src="/_next/static/chunks/framework-c3d480eb9ad30ba5.js" defer=""></script><script src="/_next/static/chunks/main-853e5508e37bbe4c.js" defer=""></script><script src="/_next/static/chunks/pages/_app-998b254a061422d4.js" defer=""></script><script src="/_next/static/chunks/pages/_error-94f0db477e27e000.js" defer=""></script><script src="/_next/static/lMO8mRWL6KkcjtN4Giq14/_buildManifest.js" defer=""></script><script src="/_next/static/lMO8mRWL6KkcjtN4Giq14/_ssgManifest.js" defer=""></script></head><body><div id="__next"><div style="font-family:system-ui,&quot;Segoe UI&quot;,Roboto,Helvetica,Arial,sans-serif,&quot;Apple Color Emoji&quot;,&quot;Segoe UI Emoji&quot;;height:100vh;text-align:center;display:flex;flex-direction:column;align-items:center;justify-content:center"><div style="line-height:48px"><style>body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}</style><h1 class="next-error-h1" style="display:inline-block;margin:0 20px 0 0;padding-right:23px;font-size:24px;font-weight:500;vertical-align:top">404</h1><div style="display:inline-block"><h2 style="font-size:14px;font-weight:400;line-height:28px">This page could not be found<!-- -->.</h2></div></div></div></div><script id="__NEXT_DATA__" type="application/json">{"props":{"pageProps":{"statusCode":404}},"page":"/_error","query":{},"buildId":"lMO8mRWL6KkcjtN4Giq14","nextExport":true,"isFallback":false,"gip":true,"scriptLoader":[]}</script></body></html>
1
+ <!DOCTYPE html><html><head><meta charSet="utf-8"/><meta name="viewport" content="width=device-width"/><title>404: This page could not be found</title><meta name="next-head-count" content="3"/><noscript data-n-css=""></noscript><script defer="" nomodule="" src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js"></script><script src="/_next/static/chunks/webpack-d1a82d85a184970b.js" defer=""></script><script src="/_next/static/chunks/framework-c3d480eb9ad30ba5.js" defer=""></script><script src="/_next/static/chunks/main-853e5508e37bbe4c.js" defer=""></script><script src="/_next/static/chunks/pages/_app-998b254a061422d4.js" defer=""></script><script src="/_next/static/chunks/pages/_error-94f0db477e27e000.js" defer=""></script><script src="/_next/static/m-VSM7eKSuIZDyA_tnwFX/_buildManifest.js" defer=""></script><script src="/_next/static/m-VSM7eKSuIZDyA_tnwFX/_ssgManifest.js" defer=""></script></head><body><div id="__next"><div style="font-family:system-ui,&quot;Segoe UI&quot;,Roboto,Helvetica,Arial,sans-serif,&quot;Apple Color Emoji&quot;,&quot;Segoe UI Emoji&quot;;height:100vh;text-align:center;display:flex;flex-direction:column;align-items:center;justify-content:center"><div style="line-height:48px"><style>body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}</style><h1 class="next-error-h1" style="display:inline-block;margin:0 20px 0 0;padding-right:23px;font-size:24px;font-weight:500;vertical-align:top">404</h1><div style="display:inline-block"><h2 style="font-size:14px;font-weight:400;line-height:28px">This page could not be found<!-- -->.</h2></div></div></div></div><script id="__NEXT_DATA__" type="application/json">{"props":{"pageProps":{"statusCode":404}},"page":"/_error","query":{},"buildId":"m-VSM7eKSuIZDyA_tnwFX","nextExport":true,"isFallback":false,"gip":true,"scriptLoader":[]}</script></body></html>
@@ -1 +1 @@
1
- <!DOCTYPE html><html><head><meta charSet="utf-8"/><meta name="viewport" content="width=device-width"/><title>404: This page could not be found</title><meta name="next-head-count" content="3"/><noscript data-n-css=""></noscript><script defer="" nomodule="" src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js"></script><script src="/_next/static/chunks/webpack-2fa22c6070dd15bc.js" defer=""></script><script src="/_next/static/chunks/framework-c3d480eb9ad30ba5.js" defer=""></script><script src="/_next/static/chunks/main-853e5508e37bbe4c.js" defer=""></script><script src="/_next/static/chunks/pages/_app-998b254a061422d4.js" defer=""></script><script src="/_next/static/chunks/pages/_error-94f0db477e27e000.js" defer=""></script><script src="/_next/static/lMO8mRWL6KkcjtN4Giq14/_buildManifest.js" defer=""></script><script src="/_next/static/lMO8mRWL6KkcjtN4Giq14/_ssgManifest.js" defer=""></script></head><body><div id="__next"><div style="font-family:system-ui,&quot;Segoe UI&quot;,Roboto,Helvetica,Arial,sans-serif,&quot;Apple Color Emoji&quot;,&quot;Segoe UI Emoji&quot;;height:100vh;text-align:center;display:flex;flex-direction:column;align-items:center;justify-content:center"><div style="line-height:48px"><style>body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}</style><h1 class="next-error-h1" style="display:inline-block;margin:0 20px 0 0;padding-right:23px;font-size:24px;font-weight:500;vertical-align:top">404</h1><div style="display:inline-block"><h2 style="font-size:14px;font-weight:400;line-height:28px">This page could not be found<!-- -->.</h2></div></div></div></div><script id="__NEXT_DATA__" type="application/json">{"props":{"pageProps":{"statusCode":404}},"page":"/_error","query":{},"buildId":"lMO8mRWL6KkcjtN4Giq14","nextExport":true,"isFallback":false,"gip":true,"scriptLoader":[]}</script></body></html>
1
+ <!DOCTYPE html><html><head><meta charSet="utf-8"/><meta name="viewport" content="width=device-width"/><title>404: This page could not be found</title><meta name="next-head-count" content="3"/><noscript data-n-css=""></noscript><script defer="" nomodule="" src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js"></script><script src="/_next/static/chunks/webpack-d1a82d85a184970b.js" defer=""></script><script src="/_next/static/chunks/framework-c3d480eb9ad30ba5.js" defer=""></script><script src="/_next/static/chunks/main-853e5508e37bbe4c.js" defer=""></script><script src="/_next/static/chunks/pages/_app-998b254a061422d4.js" defer=""></script><script src="/_next/static/chunks/pages/_error-94f0db477e27e000.js" defer=""></script><script src="/_next/static/m-VSM7eKSuIZDyA_tnwFX/_buildManifest.js" defer=""></script><script src="/_next/static/m-VSM7eKSuIZDyA_tnwFX/_ssgManifest.js" defer=""></script></head><body><div id="__next"><div style="font-family:system-ui,&quot;Segoe UI&quot;,Roboto,Helvetica,Arial,sans-serif,&quot;Apple Color Emoji&quot;,&quot;Segoe UI Emoji&quot;;height:100vh;text-align:center;display:flex;flex-direction:column;align-items:center;justify-content:center"><div style="line-height:48px"><style>body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}</style><h1 class="next-error-h1" style="display:inline-block;margin:0 20px 0 0;padding-right:23px;font-size:24px;font-weight:500;vertical-align:top">404</h1><div style="display:inline-block"><h2 style="font-size:14px;font-weight:400;line-height:28px">This page could not be found<!-- -->.</h2></div></div></div></div><script id="__NEXT_DATA__" type="application/json">{"props":{"pageProps":{"statusCode":404}},"page":"/_error","query":{},"buildId":"m-VSM7eKSuIZDyA_tnwFX","nextExport":true,"isFallback":false,"gip":true,"scriptLoader":[]}</script></body></html>
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[2],{49532:function(e,t,s){"use strict";let n,l;async function a(){if(n||(l||(l=fetch("/api/config").then(e=>e.json()).then(e=>n=e.apiBaseUrl)),await l),void 0===n)throw Error("API base URL is undefined");return n}s.d(t,{b:function(){return a}})},72002:function(e,t,s){"use strict";s.r(t),s.d(t,{default:function(){return e1}});var n=s(24004),l=s(14978),a=s(77580);let r=a.env.NEXT_PUBLIC_PROMPTFOO_REMOTE_API_BASE_URL||a.env.NEXT_PUBLIC_PROMPTFOO_BASE_URL||a.env.PROMPTFOO_REMOTE_API_BASE_URL||"https://api.promptfoo.dev",i=a.env.NEXT_PUBLIC_PROMPTFOO_BASE_URL||a.env.PROMPTFOO_REMOTE_APP_BASE_URL||"https://app.promptfoo.dev";var o=s(49532);let c=(0,l.createContext)(void 0),d=e=>{let{children:t}=e,[s,a]=(0,l.useState)(!1);return(0,l.useEffect)(()=>{let e=e=>{"Shift"===e.key&&a(!0)},t=e=>{"Shift"===e.key&&a(!1)};return window.addEventListener("keydown",e),window.addEventListener("keyup",t),()=>{window.removeEventListener("keydown",e),window.removeEventListener("keyup",t)}},[]),(0,n.jsx)(c.Provider,{value:s,children:t})};var u=s(52428),p=s(11615),h=s(12594),m=s(28891),x=s(74997),g=s(93179),j=s(47887),v=s(43205),f=s(5592),b=s(14931),y=s(75307),Z=s(17178),w=s(42610),C=s(68873),k=s(49567),S=s(40982),N=s(24362),R=s(39813),I=s(1852),P=s(57027),E=s(94941),F=s(36273),O=s(96976),T=s(8508),M=s(34304),D=s(32414),L=s(67339),A=s(65969),U=s(63147),W=s(14059),_=s(23223),z=s(26110),V=s(48931),B=s(99645),J=s(29794),H=s(22701),G=s(51956),X=s(82669),Y=s(77656),q=s(37204),$=s(79685),K=s(74595);let Q={getItem:async e=>await (0,q.U2)(e)||null,setItem:async(e,t)=>{await (0,q.t8)(e,t)},removeItem:async e=>{await (0,q.IV)(e)}},ee=(0,$.Ue)()((0,K.tJ)((e,t)=>({evalId:null,setEvalId:t=>e(()=>({evalId:t})),table:null,setTable:t=>e(()=>({table:t})),config:null,setConfig:t=>e(()=>({config:t})),maxTextLength:250,setMaxTextLength:t=>e(()=>({maxTextLength:t})),wordBreak:"break-word",setWordBreak:t=>e(()=>({wordBreak:t})),showInferenceDetails:!0,setShowInferenceDetails:t=>e(()=>({showInferenceDetails:t})),renderMarkdown:!1,setRenderMarkdown:t=>e(()=>({renderMarkdown:t})),prettifyJson:!1,setPrettifyJson:t=>e(()=>({prettifyJson:t})),showPrompts:!1,setShowPrompts:t=>e(()=>({showPrompts:t})),showPassFail:!0,setShowPassFail:t=>e(()=>({showPassFail:t}))}),{name:"ResultsViewStorage",storage:(0,K.FL)(()=>Q)}));function et(e){let{open:t,onClose:a}=e,{config:r}=ee(),i=l.useRef(null),[o,c]=l.useState(!1),[d,u]=l.useState("");l.useEffect(()=>{t&&(async()=>{let{default:e}=await Promise.resolve().then(s.bind(s,34235));u(e.dump(r))})()},[t,r]);let p=()=>{c(!1),a()};return(0,n.jsxs)(J.Z,{open:t,onClose:p,"aria-labelledby":"config-dialog-title",maxWidth:"md",fullWidth:!0,children:[(0,n.jsx)(X.Z,{id:"config-dialog-title",children:(0,n.jsxs)(k.Z,{display:"flex",justifyContent:"space-between",alignItems:"center",children:[(0,n.jsx)(U.Z,{variant:"h6",style:{flexGrow:1},children:"Config"}),(0,n.jsxs)(k.Z,{children:[(0,n.jsx)(A.Z,{title:"Copy to clipboard",children:(0,n.jsx)(Y.Z,{onClick:()=>{i.current&&(i.current.select(),document.execCommand("copy"),c(!0))},children:o?(0,n.jsx)(z.Z,{}):(0,n.jsx)(B.Z,{})})}),(0,n.jsx)(A.Z,{title:"Download .yaml",children:(0,n.jsx)(Y.Z,{onClick:()=>{let e=new Blob([d],{type:"text/yaml;charset=utf-8"}),t=URL.createObjectURL(e),s=document.createElement("a");s.href=t,s.download="config.yaml",document.body.appendChild(s),s.click(),document.body.removeChild(s),URL.revokeObjectURL(t)},children:(0,n.jsx)(V.Z,{})})})]})]})}),(0,n.jsx)(G.Z,{children:(0,n.jsx)(U.Z,{variant:"body1",component:"div",children:(0,n.jsx)("textarea",{ref:i,readOnly:!0,value:d,style:{width:"100%",minHeight:"400px",fontFamily:"monospace",border:"1px solid #ccc"}})})}),(0,n.jsx)(H.Z,{children:(0,n.jsx)(S.Z,{onClick:p,color:"primary",children:"Close"})})]})}var es=s(79715),en=s(34235),el=function(){let{table:e,config:t,evalId:s}=ee(),[a,r]=l.useState(!1),i=(e,t)=>{let s=URL.createObjectURL(e),n=document.createElement("a");n.href=s,n.download=t,document.body.appendChild(n),n.click(),document.body.removeChild(n),URL.revokeObjectURL(s)},o=()=>{r(!1)};return(0,n.jsxs)(n.Fragment,{children:[(0,n.jsxs)(O.Z,{onClick:()=>{r(!0)},children:[(0,n.jsx)(P.Z,{children:(0,n.jsx)(V.Z,{fontSize:"small"})}),(0,n.jsx)(E.Z,{children:"Download"})]}),(0,n.jsx)(J.Z,{onClose:o,open:a,children:(0,n.jsx)(G.Z,{children:(0,n.jsxs)(D.Z,{direction:"column",spacing:2,sx:{width:"100%"},children:[(0,n.jsx)(S.Z,{onClick:()=>{let e=en.default.dump(t),s=new Blob([e],{type:"text/yaml;charset=utf-8"});i(s,"promptfooconfig.yaml"),o()},startIcon:(0,n.jsx)(V.Z,{}),fullWidth:!0,sx:{justifyContent:"flex-start"},children:"Download YAML Config"}),(0,n.jsx)(S.Z,{onClick:()=>{if(!e){alert("No table data");return}let t=[],n=[...e.head.vars,...e.head.prompts.map(e=>"[".concat(e.provider,"] ").concat(e.label))];t.push(n),e.body.forEach(e=>{let s=[...e.vars,...e.outputs.map(e=>{let{pass:t,text:s}=e;return(t?"[PASS] ":"[FAIL] ")+s})];t.push(s)});let l=(0,es.P)(t),a=new Blob([l],{type:"text/csv;charset=utf-8;"});i(a,"".concat(s,"-table.csv")),o()},startIcon:(0,n.jsx)(V.Z,{}),fullWidth:!0,sx:{justifyContent:"flex-start"},children:"Download Table CSV"}),(0,n.jsx)(S.Z,{onClick:()=>{if(!e){alert("No table data");return}let t=new Blob([JSON.stringify(e,null,2)],{type:"application/json"});i(t,"".concat(s,"-table.json")),o()},startIcon:(0,n.jsx)(V.Z,{}),fullWidth:!0,sx:{justifyContent:"flex-start"},children:"Download Table JSON"}),(0,n.jsx)(S.Z,{onClick:()=>{if(!e){alert("No table data");return}let t=e.body.map((t,s)=>({chosen:t.outputs.filter(e=>e.pass).map(e=>e.text),rejected:t.outputs.filter(e=>!e.pass).map(e=>e.text),vars:t.test.vars,providers:e.head.prompts.map(e=>e.provider),prompts:e.head.prompts.map(e=>e.label||e.display||e.raw)})),n=new Blob([JSON.stringify(t,null,2)],{type:"application/json"});i(n,"".concat(s,"-dpo.json")),o()},startIcon:(0,n.jsx)(V.Z,{}),fullWidth:!0,sx:{justifyContent:"flex-start"},children:"Download DPO JSON"})]})})})]})},ea=s(69179),er=s(92715),ei=s(37803),eo=s(92863),ec=s(36001);let ed=["#fd7f6f","#7eb0d5","#b2e061","#bd7ebe","#ffb55a","#ffee65","#beb9db","#fdcce5","#8bd3c7"];function eu(e){let{table:t}=e,s=(0,l.useRef)(null),a=(0,l.useRef)(null);return(0,l.useEffect)(()=>{if(!s.current)return;a.current&&a.current.destroy();let e=t.body.flatMap(e=>e.outputs.map(e=>e.score)),n=Math.min(...e),l=(Math.ceil(Math.max(...e))-Math.floor(n))/10,r=Array.from({length:11},(e,t)=>parseFloat((Math.floor(n)+t*l).toFixed(2))),i=t.head.prompts.map((e,s)=>{let n=t.body.flatMap(e=>e.outputs[s].score),a=r.map(e=>n.filter(t=>t>=e&&t<e+l).length);return{label:"Column ".concat(s+1),data:a,backgroundColor:ed[s%ed.length]}});a.current=new ec.kL(s.current,{type:"bar",data:{labels:r,datasets:i},options:{animation:!1,plugins:{title:{display:!0,text:"Score Distribution"},legend:{display:!1},tooltip:{callbacks:{title:function(e){let t=e[0].datasetIndex;return"Column ".concat(t+1)},label:function(e){let t=e.dataIndex,s=r[t],n=r[t+1];return n?"".concat(s," <= score < ").concat(n):"".concat(s," <= score")}}}}}})},[t]),(0,n.jsx)("canvas",{ref:s,style:{maxHeight:"300px"}})}function ep(e){let{table:t}=e,s=(0,l.useRef)(null),a=(0,l.useRef)(null);return(0,l.useEffect)(()=>{if(!s.current)return;a.current&&a.current.destroy();let e=t.head.prompts.map((e,s)=>{let n=t.body.flatMap(e=>e.outputs[s]),l=n.filter(e=>e.pass).length,a=l/n.length*100;return{label:"Column ".concat(s+1),data:[a],backgroundColor:ed[s%ed.length]}});a.current=new ec.kL(s.current,{type:"bar",data:{labels:["Pass Rate (%)"],datasets:e},options:{animation:!1,plugins:{title:{display:!0,text:"Pass rate"},legend:{display:!0}}}})},[t]),(0,n.jsx)("canvas",{ref:s,style:{maxHeight:"300px"}})}function eh(e){let{table:t}=e,s=(0,l.useRef)(null),a=(0,l.useRef)(null),[r,i]=(0,l.useState)(0),[o,c]=(0,l.useState)(1),[d,u]=(0,l.useState)(!1);return(0,l.useEffect)(()=>{if(!s.current)return;a.current&&a.current.destroy();let e=t.body.flatMap(e=>e.outputs.map(e=>e.score)),n=Math.min(...e),l=Math.max(...e),i=t.body.map(e=>{let t=e.outputs[r].score,s=e.outputs[o].score;return{x:t,y:s,backgroundColor:s>t?"green":s<t?"red":"gray"}});a.current=new ec.kL(s.current,{type:"scatter",data:{datasets:[{data:i,backgroundColor:i.map(e=>e.backgroundColor)},{type:"line",data:[{x:n,y:n},{x:l,y:l}],borderColor:"gray",borderWidth:1,borderDash:[5,5],pointRadius:0}]},options:{animation:!1,plugins:{legend:{display:!1},tooltip:{callbacks:{label:function(e){let s=t.body[e.dataIndex],n=s.outputs[0].text,l=s.outputs[1].text;return n.length>30&&(n=n.substring(0,30)+"..."),l.length>30&&(l=l.substring(0,30)+"..."),"Output 1: ".concat(n,"\nOutput 2: ").concat(l)}}}},scales:{x:{title:{display:!0,text:"Prompt ".concat(r+1," Score")},ticks:{callback:function(e,t,s){let n=String(Math.round(100*Number(e)));return t===s.length-1&&(n+="%"),n}}},y:{title:{display:!0,text:"Prompt ".concat(o+1," Score")},ticks:{callback:function(e,t,s){let n=String(Math.round(100*Number(e)));return t===s.length-1&&(n+="%"),n}}}}}})},[t,r,o]),(0,n.jsxs)(n.Fragment,{children:[(0,n.jsxs)(J.Z,{open:d,onClose:()=>u(!1),children:[(0,n.jsx)(X.Z,{children:"Compare prompt outputs"}),(0,n.jsxs)(G.Z,{children:[(0,n.jsx)(R.Z,{sx:{m:1,minWidth:120},children:(0,n.jsx)(M.Z,{value:r,onChange:e=>i(Number(e.target.value)),children:t.head.prompts.map((e,t)=>(0,n.jsxs)(O.Z,{value:t,children:["Prompt ",t+1]},t))})}),(0,n.jsx)(R.Z,{sx:{m:1,minWidth:120},children:(0,n.jsx)(M.Z,{value:o,onChange:e=>c(Number(e.target.value)),children:t.head.prompts.map((e,t)=>(0,n.jsxs)(O.Z,{value:t,children:["Prompt ",t+1]},t))})})]})]}),(0,n.jsx)("canvas",{ref:s,style:{maxHeight:"300px",cursor:"pointer"},onClick:()=>u(!0)})]})}function em(e){let{table:t}=e,s=(0,l.useRef)(null),a=(0,l.useRef)(null);return(0,l.useEffect)(()=>{var e;if(!s.current)return;a.current&&a.current.destroy();let n=Object.keys((null===(e=t.head.prompts[0].metrics)||void 0===e?void 0:e.namedScores)||{}),l=t.head.prompts.map((e,s)=>{let l=n.map(s=>{var n;let l=(null===(n=e.metrics)||void 0===n?void 0:n.namedScores[s])||0,a=Math.max(...t.head.prompts.map(e=>{var t;return(null===(t=e.metrics)||void 0===t?void 0:t.namedScores[s])||0}));return l/a});return{label:"".concat(t.head.prompts[s].provider),data:l,backgroundColor:ed[s%ed.length]}});a.current=new ec.kL(s.current,{type:"bar",data:{labels:n,datasets:l},options:{scales:{x:{grid:{display:!1}},y:{ticks:{callback:function(e,t,s){let n=String(Math.round(100*Number(e)));return t===s.length-1&&(n+="%"),n}}}},plugins:{tooltip:{callbacks:{title:function(e){return e[0].dataset.label},label:function(e){let t=e.parsed.y;return"".concat(n[e.dataIndex],": ").concat((100*t).toFixed(2),"% pass rate")}}}}}})},[t]),(0,n.jsx)("canvas",{ref:s,style:{maxHeight:"300px"}})}ec.kL.register(ec.vn,ec.ST,ec.ho,ec.uw,ec.f$,ec.ZL,ec.jn,ec.od,ec.u,ec.wL);var ex=l.memo(function(e){var t;let{columnVisibility:s}=e,a=(0,eo.Z)();ec.kL.defaults.color="dark"===a.palette.mode?"#aaa":"#666";let[r,i]=(0,l.useState)(!0),{table:o}=ee();if(!o||!r||o.head.prompts.length<2)return null;let c=o.body.flatMap(e=>e.outputs.map(e=>e.score)),d=new Set(c);return 1===d.size?null:(0,n.jsx)(ea.SV,{fallback:null,children:(0,n.jsxs)(ei.Z,{style:{position:"relative",padding:a.spacing(3)},children:[(0,n.jsx)(Y.Z,{style:{position:"absolute",right:0,top:0},onClick:()=>i(!1),children:(0,n.jsx)(er.Z,{})}),(0,n.jsxs)("div",{style:{display:"flex",justifyContent:"space-between",width:"100%"},children:[(0,n.jsx)("div",{style:{width:"33%"},children:(0,n.jsx)(ep,{table:o})}),(0,n.jsx)("div",{style:{width:"33%"},children:d.size<=3&&Object.keys((null===(t=o.head.prompts[0].metrics)||void 0===t?void 0:t.namedScores)||{}).length>1?(0,n.jsx)(em,{table:o}):(0,n.jsx)(eu,{table:o})}),(0,n.jsx)("div",{style:{width:"33%"},children:(0,n.jsx)(eh,{table:o})})]})]})})}),eg=s(36128),ej=s(9317),ev=s(50459);s(55974);var ef=e=>{let{lookup:t,onSearchTextChange:s}=e;return t&&Object.keys(t).length?(0,n.jsx)("div",{className:"custom-metric-container",children:Object.entries(t).map(e=>{let[t,l]=e;return t&&void 0!==l?(0,n.jsxs)("span",{onClick:()=>s&&s("metric=".concat(t,":[^0]")),className:s?"clickable":"",children:[t,": ",l.toFixed(2)]},t):null})}):null},eb=s(35185),ey=s(65068),eZ=s(70417),ew=s(21303),eC=s(61451),ek=s(26485),eS=s(30021),eN=s(613);function eR(e){let{gradingResults:t}=e;return t?(0,n.jsxs)(k.Z,{mt:2,children:[(0,n.jsx)(U.Z,{variant:"subtitle1",children:"Assertions"}),(0,n.jsx)(eC.Z,{children:(0,n.jsxs)(ey.Z,{children:[(0,n.jsx)(ek.Z,{children:(0,n.jsxs)(eS.Z,{children:[(0,n.jsx)(ew.Z,{style:{fontWeight:"bold"},children:"Pass"}),(0,n.jsx)(ew.Z,{style:{fontWeight:"bold"},children:"Score"}),(0,n.jsx)(ew.Z,{style:{fontWeight:"bold"},children:"Type"}),(0,n.jsx)(ew.Z,{style:{fontWeight:"bold"},children:"Value"}),(0,n.jsx)(ew.Z,{style:{fontWeight:"bold"},children:"Reason"})]})}),(0,n.jsx)(eZ.Z,{children:t.map((e,t)=>{var s,l;return e?(0,n.jsxs)(eS.Z,{children:[(0,n.jsx)(ew.Z,{children:e.pass?"✅":"❌"}),(0,n.jsx)(ew.Z,{children:e.score.toFixed(2)}),(0,n.jsx)(ew.Z,{children:(null===(s=e.assertion)||void 0===s?void 0:s.type)||""}),(0,n.jsx)(ew.Z,{style:{whiteSpace:"pre-wrap"},children:(null===(l=e.assertion)||void 0===l?void 0:l.value)?String(e.assertion.value):"-"}),(0,n.jsx)(ew.Z,{style:{whiteSpace:"pre-wrap"},children:e.reason})]},t):null})})]})})]}):null}function eI(e){let{open:t,onClose:s,prompt:a,provider:r,output:i,gradingResults:o,metadata:c}=e,[d,u]=(0,l.useState)(!1);(0,l.useEffect)(()=>{u(!1)},[a]);let p=async e=>{await navigator.clipboard.writeText(e),u(!0)};return(0,n.jsxs)(J.Z,{open:t,onClose:s,fullWidth:!0,maxWidth:"lg",children:[(0,n.jsxs)(X.Z,{children:["Details",r&&": ".concat(r)]}),(0,n.jsxs)(G.Z,{children:[(0,n.jsxs)(k.Z,{mb:2,children:[(0,n.jsx)(U.Z,{variant:"subtitle1",style:{marginBottom:"1rem"},children:"Prompt"}),(0,n.jsx)(eN.u,{readOnly:!0,value:a,style:{width:"100%",padding:"0.75rem"},maxRows:20}),(0,n.jsx)(Y.Z,{onClick:()=>p(a),style:{position:"absolute",right:"10px",top:"10px"},children:d?(0,n.jsx)(z.Z,{}):(0,n.jsx)(eb.Z,{})})]}),(null==c?void 0:c.redteamFinalPrompt)&&(0,n.jsxs)(k.Z,{my:2,children:[(0,n.jsx)(U.Z,{variant:"subtitle1",style:{marginBottom:"1rem",marginTop:"1rem"},children:"Modified Prompt (Red Team)"}),(0,n.jsx)(eN.u,{readOnly:!0,maxRows:20,value:c.redteamFinalPrompt,style:{width:"100%",padding:"0.75rem"}})]}),i&&(0,n.jsxs)(k.Z,{my:2,children:[(0,n.jsx)(U.Z,{variant:"subtitle1",style:{marginBottom:"1rem",marginTop:"1rem"},children:"Output"}),(0,n.jsx)(eN.u,{readOnly:!0,maxRows:20,value:i,style:{width:"100%",padding:"0.75rem"}})]}),(0,n.jsx)(eR,{gradingResults:o})]}),(0,n.jsx)(H.Z,{children:(0,n.jsx)(S.Z,{onClick:s,children:"Close"})})]})}var eP=s(12431),eE=s(8541),eF=function(){let e=(0,eo.Z)(),[t,s]=l.useState(!1),a=()=>{s(!1)};return(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)("div",{style:{textAlign:"center",marginTop:20,marginBottom:40},children:(0,n.jsx)(S.Z,{variant:"text",color:"primary",startIcon:(0,n.jsx)(eP.Z,{}),onClick:()=>{s(!0)},children:"Generate test cases"})}),(0,n.jsxs)(J.Z,{open:t,onClose:a,children:[(0,n.jsx)(X.Z,{children:"Run on Command Line"}),(0,n.jsx)(G.Z,{children:(0,n.jsxs)(eE.Z,{children:[(0,n.jsx)("p",{children:"This feature is in beta. UI coming soon."}),(0,n.jsxs)("p",{children:["Run"," ",(0,n.jsx)(k.Z,{component:"code",sx:{backgroundColor:"dark"===e.palette.mode?"#424242":"#f0f0f0",padding:"2px 4px",borderRadius:"4px"},children:"promptfoo generate dataset"}),"to generate test cases on the command line."]})]})}),(0,n.jsx)(H.Z,{children:(0,n.jsx)(S.Z,{onClick:a,color:"primary",children:"Close"})})]})]})};function eO(e){return"string"==typeof e||"number"==typeof e?e.toString().length:Array.isArray(e)?e.reduce((e,t)=>e+eO(t),0):l.isValidElement(e)&&e.props.children?l.Children.toArray(e.props.children).reduce((e,t)=>e+eO(t),0):0}let eT=l.memo(function(e){let t,{text:s,maxLength:a}=e,[r,i]=l.useState(!0),o=function(e){let t=arguments.length>1&&void 0!==arguments[1]?arguments[1]:0;if("string"==typeof e||"number"==typeof e){let s=e.toString();return s.slice(0,a-t)}if(Array.isArray(e)){let s=[],n=t;for(let t of e){let e=eO(t);if(n+e>a){s.push(o(t,n));break}s.push(t),n+=e}return s}if(l.isValidElement(e)&&e.props.children){let s=eO(e.props.children);if(s>a-t)return l.cloneElement(e,{...e.props,children:o(e.props.children,t)})}return e};t=l.isValidElement(s)||"string"==typeof s?s:JSON.stringify(s);let c=r?o(t):t,d=eO(t)>a;return(0,n.jsxs)("div",{style:{cursor:d?"pointer":"normal"},onClick:()=>{i(!r)},children:[c,r&&eO(t)>a&&(0,n.jsx)("span",{children:"..."})]})});var eM=s(59682),eD=s(62268),eL=s(38640),eA=s.n(eL),eU=s(73490),eW=s(24495);s(16658);var e_=e=>{let{failReasons:t}=e,[s,a]=(0,l.useState)(0);return t.length<1?null:(0,n.jsxs)("div",{className:"fail-reason",children:[t.length>1&&(0,n.jsxs)("span",{className:"fail-reason-carousel-controls",children:[(0,n.jsx)(Y.Z,{onClick:()=>{a(e=>e>0?e-1:t.length-1)},children:(0,n.jsx)(eU.Z,{sx:{fontSize:12}})}),(0,n.jsxs)("span",{children:[s+1,"/",t.length]}),(0,n.jsx)(Y.Z,{onClick:()=>{a(e=>e<t.length-1?e+1:0)},children:(0,n.jsx)(eW.Z,{sx:{fontSize:12}})})]}),t[s].trim().split("\n").map((e,t)=>(0,n.jsxs)(l.Fragment,{children:[e,(0,n.jsx)("br",{})]},t))]})},ez=e=>{let{open:t,contextText:s,commentText:l,onClose:a,onSave:r,onChange:i}=e;return(0,n.jsxs)(J.Z,{open:t,onClose:a,fullWidth:!0,maxWidth:"sm",children:[(0,n.jsx)(X.Z,{children:"Edit Comment"}),(0,n.jsxs)(G.Z,{children:[(0,n.jsx)(k.Z,{sx:{backgroundColor:"#f0f0f0",padding:2,marginBottom:2},children:s}),(0,n.jsx)(L.Z,{autoFocus:!0,margin:"dense",type:"text",fullWidth:!0,multiline:!0,rows:4,value:l,onChange:e=>i(e.target.value)})]}),(0,n.jsxs)(H.Z,{children:[(0,n.jsx)(S.Z,{onClick:r,color:"primary",variant:"contained",children:"Save"}),(0,n.jsx)(S.Z,{onClick:a,color:"primary",children:"Cancel"})]})]})};let eV=()=>{let e=(0,l.useContext)(c);if(void 0===e)throw Error("useShiftKey must be used within a ShiftKeyProvider");return e};var eB=s(19116);let eJ=l.memo(function(e){var t,s,a,r,i,o,c,d,u,p,h,m,x;let g,j,v,f,b,y,{output:Z,maxTextLength:w,rowIndex:C,promptIndex:k,onRating:S,firstOutput:N,showDiffs:R,searchText:I,showStats:P}=e,{renderMarkdown:E,prettifyJson:F,showPrompts:O,showPassFail:T}=ee(),[M,D]=l.useState(!1),[L,U]=l.useState(!1),[W,_]=l.useState(null),z=e=>{_(e||null),U(!L)},[V,B]=l.useState(!1),[J,H]=l.useState((null===(t=Z.gradingResult)||void 0===t?void 0:t.comment)||""),G=()=>{B(!0)},X="string"==typeof Z.text?Z.text:JSON.stringify(Z.text),Y=[];if(!Z.pass&&X.includes("---")&&(Y=((null===(p=Z.gradingResult)||void 0===p?void 0:p.componentResults)||[]).filter(e=>!!e&&!e.pass).map(e=>e.reason),X=X.split("---").slice(1).join("---")),R&&N){let e,t="string"==typeof N.text?N.text:JSON.stringify(N.text);t.includes("---")&&(t=t.split("---").slice(1).join("---"));try{JSON.parse(t),JSON.parse(X),e=(0,eB.CT)(t,X)}catch(s){e=t.includes(". ")&&X.includes(". ")?(0,eB.SY)(t,X):(0,eB.NV)(t,X)}g=(0,n.jsx)(n.Fragment,{children:e.map((e,t)=>e.added?(0,n.jsx)("ins",{children:e.value},t):e.removed?(0,n.jsx)("del",{children:e.value},t):(0,n.jsx)("span",{children:e.value},t))})}if(I)try{let e;let t=RegExp(I,"gi"),s=[];for(;null!==(e=t.exec(X));)s.push({start:e.index,end:t.lastIndex});g=(0,n.jsx)(n.Fragment,{children:s.length>0?(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)("span",{children:X.substring(0,s[0].start)},"text-before"),s.map((e,t)=>(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)("span",{className:"search-highlight",children:X.substring(e.start,e.end)},"match-"+t),(0,n.jsx)("span",{children:X.substring(e.end,s[t+1]?s[t+1].start:X.length)},"text-after-"+t)]}))]}):(0,n.jsx)("span",{children:X},"no-match")})}catch(e){console.error("Invalid regular expression:",e.message)}else if(E)g=(0,n.jsx)(ev.U,{components:{img:e=>{let{src:t,alt:s}=e;return(0,n.jsx)("img",{loading:"lazy",src:t,alt:s,onClick:()=>z(t),style:{cursor:"pointer"}})}},children:X});else if(F)try{g=(0,n.jsx)("pre",{children:JSON.stringify(JSON.parse(X),null,2)})}catch(e){}let q=l.useCallback(e=>{var t;S(e,void 0,null===(t=Z.gradingResult)||void 0===t?void 0:t.comment)},[S,null===(s=Z.gradingResult)||void 0===s?void 0:s.comment]),$=l.useCallback(()=>{let e=prompt("Set test score (0.0 - 1.0):",String(Z.score));if(null!==e){let s=parseFloat(e);if(!isNaN(s)&&s>=0&&s<=1){var t;S(void 0,s,null===(t=Z.gradingResult)||void 0===t?void 0:t.comment)}else alert("Invalid score. Please enter a value between 0.0 and 1.0.")}},[S,Z.score,null===(a=Z.gradingResult)||void 0===a?void 0:a.comment]),[K,Q]=l.useState(!1),et=l.useCallback(()=>{navigator.clipboard.writeText(Z.text),Q(!0)},[Z.text]);if(Z.latencyMs&&(v=(0,n.jsxs)("span",{children:[Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(Z.latencyMs)," ms"]})),null===(r=Z.tokenUsage)||void 0===r?void 0:r.completion){let e=Z.tokenUsage.completion/(Z.latencyMs/1e3);f=(0,n.jsx)("span",{children:Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(e)})}if(Z.cost&&(b=(0,n.jsxs)("span",{children:["$",Z.cost.toPrecision(2)]})),null===(i=Z.tokenUsage)||void 0===i?void 0:i.cached)j=(0,n.jsxs)("span",{children:[Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(Z.tokenUsage.cached)," ","(cached)"]});else if(null===(o=Z.tokenUsage)||void 0===o?void 0:o.total){let e=Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(null!==(h=Z.tokenUsage.prompt)&&void 0!==h?h:0),t=Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(null!==(m=Z.tokenUsage.completion)&&void 0!==m?m:0),s=Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(Z.tokenUsage.total);j=(0,n.jsx)(A.Z,{title:"".concat(e," prompt tokens + ").concat(t," completion tokens = ").concat(s," total"),children:(0,n.jsxs)("span",{children:[s,("0"!==e||"0"!==t)&&" (".concat(e,"+").concat(t,")")]})})}let es=(null===(c=Z.gradingResult)||void 0===c?void 0:c.comment)&&"!highlight"!==Z.gradingResult.comment?(0,n.jsx)("div",{className:"comment",onClick:G,children:Z.gradingResult.comment}):null,en=P?(0,n.jsxs)("div",{className:"cell-detail",children:[j&&(0,n.jsxs)("div",{className:"stat-item",children:[(0,n.jsx)("strong",{children:"Tokens:"})," ",j]}),v&&(0,n.jsxs)("div",{className:"stat-item",children:[(0,n.jsx)("strong",{children:"Latency:"})," ",v]}),f&&(0,n.jsxs)("div",{className:"stat-item",children:[(0,n.jsx)("strong",{children:"Tokens/Sec:"})," ",f]}),b&&(0,n.jsxs)("div",{className:"stat-item",children:[(0,n.jsx)("strong",{children:"Cost:"})," ",b]})]}):null,el=eV(),ea=(0,n.jsxs)("div",{className:"cell-actions",children:[el&&(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)("span",{className:"action",onClick:et,onMouseDown:e=>e.preventDefault(),children:(0,n.jsx)(A.Z,{title:"Copy output to clipboard",children:(0,n.jsx)("span",{children:K?"✅":"\uD83D\uDCCB"})})}),(0,n.jsx)("span",{className:"action",onClick:()=>{let e;J.startsWith("!highlight")?S(void 0,void 0,e=J.slice(10).trim()):S(void 0,void 0,e=("!highlight "+J).trim()),H(e)},onMouseDown:e=>e.preventDefault(),children:(0,n.jsx)(A.Z,{title:"Toggle test highlight",children:(0,n.jsx)("span",{children:"\uD83C\uDF1F"})})})]}),Z.prompt&&(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)("span",{className:"action",onClick:()=>{D(!0)},children:(0,n.jsx)(A.Z,{title:"View ouput and test details",children:(0,n.jsx)("span",{children:"\uD83D\uDD0E"})})}),(0,n.jsx)(eI,{open:M,onClose:()=>{D(!1)},prompt:Z.prompt,provider:Z.provider,gradingResults:null===(d=Z.gradingResult)||void 0===d?void 0:d.componentResults,output:X,metadata:Z.metadata})]}),(0,n.jsx)("span",{className:"action",onClick:()=>q(!0),children:(0,n.jsx)(A.Z,{title:"Mark test passed (score 1.0)",children:(0,n.jsx)("span",{children:"\uD83D\uDC4D"})})}),(0,n.jsx)("span",{className:"action",onClick:()=>q(!1),children:(0,n.jsx)(A.Z,{title:"Mark test failed (score 0.0)",children:(0,n.jsx)("span",{children:"\uD83D\uDC4E"})})}),(0,n.jsx)("span",{className:"action",onClick:$,children:(0,n.jsx)(A.Z,{title:"Set test score",children:(0,n.jsx)("span",{children:"\uD83D\uDD22"})})}),(0,n.jsx)("span",{className:"action",onClick:G,children:(0,n.jsx)(A.Z,{title:"Edit comment",children:(0,n.jsx)("span",{children:"✏️"})})})]}),er={};(null===(u=Z.gradingResult)||void 0===u?void 0:u.comment)==="!highlight"&&(er.backgroundColor="#ffffeb");let ei=0,eo=0,ec=Z.gradingResult;if(ec?ec.componentResults?ec.componentResults.forEach(e=>{(null==e?void 0:e.pass)?ei++:eo++}):(ei=ec.pass?1:0,eo=ec.pass?0:1):Z.pass?ei=1:Z.pass||(eo=1),1===eo&&1===ei)y=(0,n.jsxs)(n.Fragment,{children:["".concat(eo," FAIL")," ","".concat(ei," PASS")]});else{let e="";eo>1||ei>1&&eo>0?e="".concat(eo," FAIL"):1===eo&&(e="FAIL");let t="";ei>1||eo>1&&ei>0?t="".concat(ei," PASS"):1===ei&&0===eo&&(t="PASS");let s=e&&t?" ":"";y=(0,n.jsxs)(n.Fragment,{children:[e,s,t]})}let ed=null===(x=Z.score)||0===x||1===x?"":"(".concat(x.toFixed(2),")");return(0,n.jsxs)("div",{className:"cell",style:er,children:[T&&(0,n.jsx)(n.Fragment,{children:Z.pass?(0,n.jsx)(n.Fragment,{children:(0,n.jsxs)("div",{className:"status pass",children:[(0,n.jsxs)("div",{className:"pill",children:[y,ed&&(0,n.jsxs)("span",{className:"score",children:[" ",ed]})]}),(0,n.jsx)(ef,{lookup:Z.namedScores})]})}):(0,n.jsx)(n.Fragment,{children:(0,n.jsxs)("div",{className:"status fail",children:[(0,n.jsxs)("div",{className:"pill",children:[y,ed&&(0,n.jsxs)("span",{className:"score",children:[" ",ed]})]}),(0,n.jsx)(ef,{lookup:Z.namedScores}),(0,n.jsx)("span",{className:"fail-reason",children:(0,n.jsx)(e_,{failReasons:Y})})]})})}),O&&N.prompt&&(0,n.jsxs)("div",{className:"prompt",children:[(0,n.jsx)("span",{className:"pill",children:"Prompt"}),Z.prompt]}),(0,n.jsx)(eT,{text:g||X,maxLength:w}),es,en,ea,L&&W&&(0,n.jsx)("div",{className:"lightbox",onClick:()=>z(),children:(0,n.jsx)("img",{src:W,alt:"Lightbox"})}),(0,n.jsx)(ez,{open:V,contextText:Z.text,commentText:J,onClose:()=>{B(!1)},onSave:()=>{S(void 0,void 0,J),B(!1)},onChange:H})]})});function eH(e){let{text:t,maxLength:s,expandedText:a,resourceId:r,className:i}=e,[o,c]=l.useState(!1);return(0,n.jsxs)("div",{className:"".concat(i||""),children:[(0,n.jsx)(eT,{text:t,maxLength:s}),a&&(0,n.jsxs)(n.Fragment,{children:[(0,n.jsx)(A.Z,{title:"View prompt",children:(0,n.jsx)("span",{className:"action",onClick:()=>{c(!0)},children:"\uD83D\uDD0E"})}),(0,n.jsx)(eI,{open:o,onClose:()=>{c(!1)},prompt:a}),r&&(0,n.jsx)(A.Z,{title:"View other evals and datasets for this prompt",children:(0,n.jsx)("span",{className:"action",children:(0,n.jsx)(eA(),{href:"/prompts/?id=".concat(r),target:"_blank",children:(0,n.jsx)(eM.Z,{fontSize:"small"})})})})]})]})}s(93207);var eG=l.memo(function(e){let{maxTextLength:t,columnVisibility:s,wordBreak:a,filterMode:r,failureFilter:i,searchText:c,showStats:d,onFailureFilterToggle:u,onSearchTextChange:p}=e,{evalId:h,table:m,setTable:x}=ee();(0,g.Z)(m,"Table should be defined");let{head:j,body:v}=m,f=l.useCallback(async(e,t,s,n,l)=>{var a,r;let i=[...v],c={...i[e]},d=[...c.outputs],u=null!=s?s:d[t].pass,p=void 0===n?s?1:0:n||0;d[t].pass=u,d[t].score=p;let m=(null===(a=d[t].gradingResult)||void 0===a?void 0:a.componentResults)||[];if(void 0!==s){let e=m.findIndex(e=>{var t;return(null===(t=e.assertion)||void 0===t?void 0:t.type)==="human"}),t={pass:u,score:p,reason:"Manual result (overrides all other grading results)",comment:l,assertion:{type:"human"}};-1!==e?m[e]=t:m.push(t)}let g={...d[t].gradingResult||{},pass:u,score:p,reason:"Manual result (overrides all other grading results)",comment:l,assertion:(null===(r=d[t].gradingResult)||void 0===r?void 0:r.assertion)||null,componentResults:m};d[t].gradingResult=g,c.outputs=d,i[e]=c;let f={head:j,body:i};x(f);try{let e=await fetch("".concat(await (0,o.b)(),"/api/eval/").concat(h),{method:"PATCH",headers:{"Content-Type":"application/json"},body:JSON.stringify({table:f})});if(!e.ok)throw Error("Network response was not ok")}catch(e){console.error("Failed to update table:",e)}},[v,j,x,h]),b=Object.keys(s).length>0,y=l.useMemo(()=>{try{return RegExp(c,"i")}catch(e){return console.error("Invalid regular expression:",e.message),null}},[c]),Z=l.useMemo(()=>{try{return v.map((e,t)=>({...e,outputs:e.outputs.map((e,s)=>({...e,originalRowIndex:t,originalPromptIndex:s}))})).filter(e=>{let t=!0;return"failures"===r?t=e.outputs.some((e,t)=>{let n="Prompt ".concat(t+1);return i[n]&&!e.pass&&(!b||s[n])}):"different"===r?t=!e.outputs.every(t=>t.text===e.outputs[0].text):"highlights"===r&&(console.log(e.outputs[0].text),t=e.outputs.some(e=>{var t,s;return null===(s=e.gradingResult)||void 0===s?void 0:null===(t=s.comment)||void 0===t?void 0:t.startsWith("!highlight")})),!!t&&(!c||!y||e.outputs.some(t=>{var s,n;let l=e.vars.map(e=>"var=".concat(e)).join(" "),a="".concat(t.text," ").concat(Object.keys(t.namedScores).map(e=>"metric=".concat(e,":").concat(t.namedScores[e])).join(" ")," ").concat((null===(s=t.gradingResult)||void 0===s?void 0:s.reason)||""," ").concat((null===(n=t.gradingResult)||void 0===n?void 0:n.comment)||""),r="".concat(l," ").concat(a);return y.test(r)}))})}catch(e){return console.error("Invalid regular expression:",e.message),v}},[v,i,r,c,s,b,y]);l.useEffect(()=>{B(e=>({...e,pageIndex:0}))},[i,r,c]);let w=l.useMemo(()=>j.prompts.map((e,t)=>v.reduce((e,s)=>e+(s.outputs[t].pass?1:0),0)),[j.prompts,v]),C=l.useMemo(()=>j.prompts.map((e,t)=>v.reduce((e,s)=>{var n,l;return e+((null===(l=s.outputs[t].gradingResult)||void 0===l?void 0:null===(n=l.componentResults)||void 0===n?void 0:n.length)||0)},0)),[j.prompts,v]),R=l.useMemo(()=>j.prompts.map((e,t)=>v.reduce((e,s)=>{var n;let l=null===(n=s.outputs[t].gradingResult)||void 0===n?void 0:n.componentResults;return e+(l?l.filter(e=>null==e?void 0:e.pass).length:0)},0)),[j.prompts,v]),I=l.useMemo(()=>w.reduce((e,t,s,n)=>t>n[e]?s:e,0),[w]),P=w[I],E=l.useMemo(()=>(0,eg.Cl)(),[]),{renderMarkdown:F}=ee(),T=l.useMemo(()=>j.vars.length>0?[E.group({id:"vars",header:()=>(0,n.jsx)("span",{className:"font-bold",children:"Variables"}),columns:j.vars.map((e,s)=>E.accessor(e=>e.vars[s],{id:"Variable ".concat(s+1),header:()=>(0,n.jsx)(eH,{text:e,maxLength:t,className:"font-bold"}),cell:e=>{let s=e.getValue();return(0,n.jsx)("div",{className:"cell",children:F?(0,n.jsx)(ev.U,{children:s}):(0,n.jsx)(eT,{text:s,maxLength:t})})},size:50}))})]:[],[E,j.vars,t,F]),D=l.useCallback((e,t)=>Z[e].outputs[t],[Z]),A=l.useCallback(e=>Z[e].outputs[0],[Z]),W=l.useMemo(()=>[E.group({id:"prompts",header:()=>(0,n.jsx)("span",{className:"font-bold",children:"Outputs"}),columns:j.prompts.map((e,s)=>E.accessor(e=>(function(e){if("string"==typeof e){let t=e.startsWith("[PASS]"),s=e;return e.startsWith("[PASS]")?s=s.slice(6):e.startsWith("[FAIL]")&&(s=s.slice(6)),{text:s,pass:t,score:t?1:0}}return e})(e.outputs[s]),{id:"Prompt ".concat(s+1),header:()=>{var l,a,o,c,h,m,x,g;let j=w[s]&&v.length?(w[s]/v.length*100).toFixed(2):"0.00",f=w[s]===P&&0!==P,b="Prompt ".concat(s+1),y=i[b]||!1,Z=d?(0,n.jsxs)("div",{className:"prompt-detail",children:[C[s]?(0,n.jsxs)("div",{children:[(0,n.jsx)("strong",{children:"Asserts:"})," ",R[s],"/",C[s]," passed"]}):null,(null===(l=e.metrics)||void 0===l?void 0:l.totalLatencyMs)?(0,n.jsxs)("div",{children:[(0,n.jsx)("strong",{children:"Avg Latency:"})," ",Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(e.metrics.totalLatencyMs/v.length)," ","ms"]}):null,(null===(o=e.metrics)||void 0===o?void 0:null===(a=o.tokenUsage)||void 0===a?void 0:a.total)?(0,n.jsxs)("div",{children:[(0,n.jsx)("strong",{children:"Avg Tokens:"})," ",Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(e.metrics.tokenUsage.total/v.length)]}):null,(null===(c=e.metrics)||void 0===c?void 0:c.totalLatencyMs)&&(null===(m=e.metrics)||void 0===m?void 0:null===(h=m.tokenUsage)||void 0===h?void 0:h.completion)?(0,n.jsxs)("div",{children:[(0,n.jsx)("strong",{children:"Tokens/Sec:"})," ",Intl.NumberFormat(void 0,{maximumFractionDigits:0}).format(e.metrics.tokenUsage.completion/(e.metrics.totalLatencyMs/1e3))]}):null,(null===(x=e.metrics)||void 0===x?void 0:x.cost)?(0,n.jsxs)("div",{children:[(0,n.jsx)("strong",{children:"Cost:"})," $",e.metrics.cost.toPrecision(2)]}):null]}):null,k=e.provider?e.provider.split(":"):[],S=k.length>1?(0,n.jsxs)(n.Fragment,{children:[k[0],":",(0,n.jsx)("strong",{children:k.slice(1).join(":")})]}):(0,n.jsx)("strong",{children:e.provider});return(0,n.jsxs)("div",{className:"output-header",children:[(0,n.jsxs)("div",{className:"pills",children:[e.provider?(0,n.jsx)("div",{className:"provider",children:S}):null,(0,n.jsx)("div",{className:"summary",children:(0,n.jsxs)("div",{className:"highlight ".concat(f?"success":""),children:[(0,n.jsxs)("strong",{children:[j,"% passing"]})," (",w[s],"/",v.length," cases)"]})}),(null===(g=e.metrics)||void 0===g?void 0:g.namedScores)&&Object.keys(e.metrics.namedScores).length>0?(0,n.jsx)(ef,{lookup:e.metrics.namedScores,onSearchTextChange:p}):null]}),(0,n.jsx)(eH,{className:"prompt-container",text:e.label||e.display||e.raw,expandedText:e.raw,maxLength:t,resourceId:e.id}),Z,"failures"===r&&(0,n.jsx)(eD.Z,{sx:{"& .MuiFormControlLabel-label":{fontSize:"0.75rem"}},control:(0,n.jsx)(N.Z,{checked:y,onChange:e=>u(b,e.target.checked)}),label:"Show failures"})]})},cell:e=>{var l,a;let i=D(e.row.index,s);return(0,n.jsx)(eJ,{output:i,maxTextLength:t,rowIndex:e.row.index,promptIndex:s,onRating:f.bind(null,null!==(l=i.originalRowIndex)&&void 0!==l?l:e.row.index,null!==(a=i.originalPromptIndex)&&void 0!==a?a:s),firstOutput:A(e.row.index),showDiffs:"different"===r,searchText:c,showStats:d})}}))})],[E,j.prompts,w,v.length,P,i,d,C,R,t,u,r,c,D,A,f,p]),_=l.useMemo(()=>{let e=v.some(e=>e.description);return e?{accessorFn:e=>e.description||"",id:"description",header:()=>(0,n.jsx)("span",{className:"font-bold",children:"Description"}),cell:e=>(0,n.jsx)("div",{className:"cell",children:(0,n.jsx)(eT,{text:String(e.getValue()),maxLength:t})}),size:50}:null},[v,t]),z=l.useMemo(()=>{let e=[];return _&&e.push(_),e.push(...T,...W),e},[_,T,W]),[V,B]=l.useState({pageIndex:0,pageSize:50}),J=(0,ej.b7)({data:Z,columns:z,columnResizeMode:"onChange",getCoreRowModel:(0,eg.sC)(),getPaginationRowModel:(0,eg.G_)(),state:{columnVisibility:s,pagination:V}});return(0,n.jsxs)("div",{children:[(0,n.jsxs)("table",{className:"results-table firefox-fix ".concat(t<=25?"compact":""),style:{wordBreak:a},children:[(0,n.jsx)("thead",{children:J.getHeaderGroups().map(e=>(0,n.jsx)("tr",{className:"header",children:e.headers.map(e=>(0,n.jsxs)("th",{colSpan:e.colSpan,style:{width:e.getSize()},children:[e.isPlaceholder?null:(0,ej.ie)(e.column.columnDef.header,e.getContext()),(0,n.jsx)("div",{onMouseDown:e.getResizeHandler(),onTouchStart:e.getResizeHandler(),className:"resizer ".concat(e.column.getIsResizing()?"isResizing":"")})]},e.id))},e.id))}),(0,n.jsx)("tbody",{children:J.getRowModel().rows.map((e,t)=>{let s=!1;return(0,n.jsx)("tr",{children:e.getVisibleCells().map(e=>{let l=e.column.id.startsWith("Variable")||"description"===e.column.id,a=!l&&!s;return a&&(s=!0),(0,n.jsx)("td",{style:{width:e.column.getSize()},className:"".concat(l?"variable":""," ").concat(0!==t||l?"":"first-prompt-row"," ").concat(a?"first-prompt-col":""),children:(0,ej.ie)(e.column.columnDef.cell,e.getContext())},e.id)})},e.id)})})]}),J.getPageCount()>1&&(0,n.jsxs)(k.Z,{className:"pagination",sx:{display:"flex",alignItems:"center",gap:2},children:[(0,n.jsx)(S.Z,{onClick:()=>{B(e=>({...e,pageIndex:Math.max(e.pageIndex-1,0)})),window.scrollTo(0,0)},disabled:0===J.getState().pagination.pageIndex,variant:"contained",children:"Previous"}),(0,n.jsxs)(U.Z,{component:"span",sx:{display:"flex",alignItems:"center",gap:1},children:["Page",(0,n.jsx)(L.Z,{size:"small",type:"number",value:J.getState().pagination.pageIndex+1,onChange:e=>{let t=e.target.value?Number(e.target.value)-1:0;B(e=>({...e,pageIndex:Math.min(Math.max(t,0),J.getPageCount()-1)}))},InputProps:{style:{width:"60px",textAlign:"center"}},variant:"outlined"}),(0,n.jsxs)("span",{children:["of ",J.getPageCount()]})]}),(0,n.jsx)(S.Z,{onClick:()=>{B(e=>({...e,pageIndex:Math.min(e.pageIndex+1,J.getPageCount()-1)})),window.scrollTo(0,0)},disabled:J.getState().pagination.pageIndex+1>=J.getPageCount(),variant:"contained",children:"Next"}),(0,n.jsxs)(U.Z,{component:"span",sx:{display:"flex",alignItems:"center",gap:1},children:[(0,n.jsxs)(M.Z,{value:V.pageSize,onChange:e=>{B({pageIndex:0,pageSize:Number(e.target.value)}),window.scrollTo(0,0)},displayEmpty:!0,inputProps:{"aria-label":"Results per page"},size:"small",sx:{m:1,minWidth:80},children:[(0,n.jsx)(O.Z,{value:10,children:"10"}),(0,n.jsx)(O.Z,{value:50,children:"50"}),(0,n.jsx)(O.Z,{value:100,children:"100"}),(0,n.jsx)(O.Z,{value:500,children:"500"}),(0,n.jsx)(O.Z,{value:1e3,children:"1000"})]}),(0,n.jsx)("span",{children:"results per page"})]})]}),(0,n.jsx)(eF,{})]})}),eX=s(22695),eY=e=>{let{open:t,onClose:s}=e,{maxTextLength:l,setMaxTextLength:a,wordBreak:r,setWordBreak:i,showInferenceDetails:o,setShowInferenceDetails:c,renderMarkdown:d,setRenderMarkdown:u,prettifyJson:p,setPrettifyJson:h,showPrompts:m,setShowPrompts:x,showPassFail:g,setShowPassFail:j}=ee();return(0,n.jsxs)(J.Z,{open:t,onClose:s,fullWidth:!0,maxWidth:"sm",children:[(0,n.jsx)(X.Z,{children:"Table View Settings"}),(0,n.jsxs)(G.Z,{children:[(0,n.jsx)(k.Z,{children:(0,n.jsx)(A.Z,{title:"Forcing line breaks makes it easier to adjust column widths to your liking",placement:"right",children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:"break-all"===r,onChange:e=>i(e.target.checked?"break-all":"break-word")}),label:"Force line breaks"})})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:d,onChange:e=>u(e.target.checked)}),label:"Render model outputs as Markdown"})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:p,onChange:e=>h(e.target.checked)}),label:"Prettify JSON outputs"})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(A.Z,{title:"Show the final prompt that produced the output in each cell.",placement:"right",children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:m,onChange:e=>x(e.target.checked)}),label:"Show full prompt in output cell"})})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(A.Z,{title:"Show pass/fail status for each output.",placement:"right",children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:g,onChange:e=>j(e.target.checked)}),label:"Show pass/fail status"})})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(A.Z,{title:"Show detailed inference statistics such as latency, tokens used, cost, etc.",placement:"right",children:(0,n.jsx)(eD.Z,{control:(0,n.jsx)(N.Z,{checked:o,onChange:e=>c(e.target.checked)}),label:"Show inference details"})})}),(0,n.jsxs)(k.Z,{maxWidth:"sm",children:[(0,n.jsxs)(U.Z,{mt:2,children:["Max text length: ",l]}),(0,n.jsx)(eX.ZP,{min:25,max:1e3,value:l,onChange:(e,t)=>a(t)})]})]}),(0,n.jsx)(H.Z,{children:(0,n.jsx)(S.Z,{onClick:s,children:"Close"})})]})},eq=e=>{let{open:t,onClose:s,shareUrl:a}=e,r=(0,l.useRef)(null),[i,o]=(0,l.useState)(!1),c=()=>{s(),o(!1)};return(0,n.jsxs)(J.Z,{open:t,onClose:c,PaperProps:{style:{minWidth:"min(660px, 100%)"}},children:[(0,n.jsx)(X.Z,{children:"Your eval is ready to share"}),(0,n.jsxs)(G.Z,{children:[(0,n.jsx)(L.Z,{inputRef:r,value:a,fullWidth:!0,InputProps:{readOnly:!0,endAdornment:(0,n.jsx)(Y.Z,{onClick:()=>{r.current&&(r.current.select(),document.execCommand("copy"),o(!0))},children:i?(0,n.jsx)(z.Z,{}):(0,n.jsx)(B.Z,{})})}}),(0,n.jsx)(eE.Z,{sx:{fontSize:"0.75rem"},children:"Shared URLs are deleted after 2 weeks."})]}),(0,n.jsx)(H.Z,{children:(0,n.jsx)(S.Z,{onClick:c,color:"primary",children:"Close"})})]})};s(58022);let e$=(0,W.Z)(D.Z)(e=>{let{theme:t}=e;return{maxWidth:"100%",flexWrap:"wrap",[t.breakpoints.down("sm")]:{flexDirection:"column"}}});function eK(e){var t;let{recentEvals:s,onRecentEvalSelected:a,defaultEvalId:c}=e,d=(0,m.useRouter)(),u=(0,m.useSearchParams)(),{table:h,config:x,setConfig:D,maxTextLength:W,wordBreak:z,showInferenceDetails:V,evalId:B}=ee(),{setStateFromConfig:J}=(0,j.o)(),[H,G]=l.useState((null==u?void 0:u.get("search"))||""),[X]=(0,_.Nr)(H,1e3),Y=e=>{G(e)},[q,$]=l.useState({}),K=l.useCallback((e,t)=>{$(s=>({...s,[e]:t}))},[$]),[Q,es]=l.useState("all"),[en,ea]=l.useState(!1),[er,ei]=l.useState(""),[eo,ec]=l.useState(!1),ed=async()=>{ec(!0);try{let e=await fetch("".concat(r,"/api/eval"),{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({data:{version:2,createdAt:new Date().toISOString(),results:{table:h},config:x}})}),{id:t}=await e.json(),s="".concat(i,"/eval/").concat(t);ei(s),ea(!0)}catch(e){alert("Sorry, something went wrong.")}finally{ec(!1)}},[eu,ep]=l.useState(!1),[eh,em]=l.useState(!1);(0,g.Z)(h,"Table data must be loaded before rendering ResultsView");let{head:eg}=h,ej=async()=>{(0,g.Z)(x,"Config must be loaded before clicking its description");let e=window.prompt("Enter new description:",x.description);if(null!==e&&e!==x.description){let t={...x,description:e};try{let e=await fetch("".concat(await (0,o.b)(),"/api/eval/").concat(B),{method:"PATCH",headers:{"Content-Type":"application/json"},body:JSON.stringify({config:t})});if(!e.ok)throw Error("Network response was not ok");D(t)}catch(e){console.error("Failed to update table:",e)}}},ev=async()=>{if(window.confirm("Are you sure you want to delete this evaluation?"))try{let e=await fetch("".concat(await (0,o.b)(),"/api/eval/").concat(B),{method:"DELETE"});if(!e.ok)throw Error("Network response was not ok");d.push("/")}catch(e){console.error("Failed to delete evaluation:",e),alert("Failed to delete evaluation")}},ef=l.useMemo(()=>[...eg.vars.map((e,t)=>({value:"Variable ".concat(t+1),label:"Var ".concat(t+1,": ").concat(eg.vars[t].length>100?eg.vars[t].slice(0,97)+"...":eg.vars[t]),group:"Variables"})),...eg.prompts.map((e,t)=>{let s=eg.prompts[t],n=s.label||s.display||s.raw;return{value:"Prompt ".concat(t+1),label:"Prompt ".concat(t+1,": ").concat(n.length>100?n.slice(0,97)+"...":n),group:"Prompts"}})],[eg.vars,eg.prompts]),[eb,ey]=l.useState({}),[eZ,ew]=l.useState(ef.map(e=>e.value)),[eC,ek]=l.useState(null);return(0,n.jsxs)("div",{style:{marginLeft:"1rem",marginRight:"1rem"},children:[(0,n.jsx)(k.Z,{mb:2,sx:{display:"flex",alignItems:"center"},children:(0,n.jsxs)(U.Z,{variant:"h5",sx:{flexGrow:1},children:[(0,n.jsx)("span",{className:"description",onClick:ej,children:(null==x?void 0:x.description)||B})," ",(null==x?void 0:x.description)&&(0,n.jsx)("span",{className:"description-filepath",children:B})]})}),(0,n.jsx)(k.Z,{py:"md",children:(0,n.jsxs)(e$,{direction:"row",spacing:4,alignItems:"center",children:[(0,n.jsx)(k.Z,{children:s&&s.length>0&&(0,n.jsx)(R.Z,{sx:{m:1,minWidth:200},size:"small",children:(0,n.jsx)(C.Z,{size:"small",options:s,renderOption:(e,t)=>(0,l.createElement)("li",{...e,key:t.id},t.label),style:{width:350},renderInput:e=>(0,n.jsx)(L.Z,{...e,label:"Eval run",variant:"outlined"}),defaultValue:s.find(e=>e.id===c)||s[0],onChange:(e,t)=>{t&&t.id&&a(t.id)},disableClearable:!0})})}),(0,n.jsx)(k.Z,{children:(0,n.jsxs)(R.Z,{sx:{m:1,minWidth:200,maxWidth:350},size:"small",children:[(0,n.jsx)(I.Z,{id:"visible-columns-label",children:"Columns"}),(0,n.jsx)(M.Z,{labelId:"visible-columns-label",id:"visible-columns",multiple:!0,value:eZ,onChange:e=>{let{target:{value:t}}=e;ew("string"==typeof t?t.split(","):t);let s=[...eg.vars.map((e,t)=>"Variable ".concat(t+1)),...eg.prompts.map((e,t)=>"Prompt ".concat(t+1))],n={};s.forEach(e=>{n[e]=("string"==typeof t?t.split(","):t).includes(e)}),ey(n)},input:(0,n.jsx)(T.Z,{label:"Visible columns"}),renderValue:e=>e.join(", "),children:ef.map(e=>(0,n.jsxs)(O.Z,{dense:!0,value:e.value,children:[(0,n.jsx)(N.Z,{checked:eZ.indexOf(e.value)>-1}),(0,n.jsx)(E.Z,{primary:e.label})]},e.value))})]})}),(0,n.jsx)(k.Z,{children:(0,n.jsxs)(R.Z,{sx:{minWidth:180},size:"small",children:[(0,n.jsx)(I.Z,{id:"failure-filter-mode-label",children:"Display"}),(0,n.jsxs)(M.Z,{labelId:"filter-mode-label",id:"filter-mode",value:Q,onChange:e=>{let t=e.target.value;es(t);let s={};eg.prompts.forEach((e,n)=>{s["Prompt ".concat(n+1)]="failures"===t}),$(s)},label:"Filter",children:[(0,n.jsx)(O.Z,{value:"all",children:"Show all results"}),(0,n.jsx)(O.Z,{value:"failures",children:"Show failures only"}),(0,n.jsx)(O.Z,{value:"different",children:"Show different only"}),(0,n.jsx)(O.Z,{value:"highlights",children:"Show highlights only"})]})]})}),(0,n.jsx)(k.Z,{children:(0,n.jsx)(L.Z,{sx:{minWidth:180},size:"small",label:"Search",placeholder:"Text or regex",value:H,onChange:e=>Y(e.target.value)})}),(0,n.jsx)(k.Z,{flexGrow:1}),(0,n.jsx)(k.Z,{display:"flex",justifyContent:"flex-end",children:(0,n.jsxs)(e$,{direction:"row",spacing:2,children:[(0,n.jsx)(S.Z,{color:"primary",onClick:e=>{ek(e.currentTarget)},startIcon:(0,n.jsx)(v.Z,{}),children:"Eval actions"}),x&&(0,n.jsxs)(F.Z,{id:"eval-actions-menu",anchorEl:eC,keepMounted:!0,open:!!eC,onClose:()=>{ek(null)},children:[(0,n.jsx)(A.Z,{title:"View the configuration that defines this eval",placement:"left",children:(0,n.jsxs)(O.Z,{onClick:()=>ep(!0),children:[(0,n.jsx)(P.Z,{children:(0,n.jsx)(w.Z,{fontSize:"small"})}),"View YAML"]})}),(0,n.jsx)(A.Z,{title:"Edit this eval in the web UI",placement:"left",children:(0,n.jsxs)(O.Z,{onClick:()=>{J(x),d.push("/setup/")},children:[(0,n.jsx)(P.Z,{children:(0,n.jsx)(b.Z,{fontSize:"small"})}),"Edit Eval"]})}),(0,n.jsx)(el,{}),(null==x?void 0:x.sharing)&&(0,n.jsx)(A.Z,{title:"Generate a unique URL that others can access",placement:"left",children:(0,n.jsxs)(O.Z,{onClick:ed,disabled:eo,children:[(0,n.jsx)(P.Z,{children:eo?(0,n.jsx)(p.Z,{size:16}):(0,n.jsx)(Z.Z,{fontSize:"small"})}),"Share"]})}),(0,n.jsx)(A.Z,{title:"Delete this eval",placement:"left",children:(0,n.jsxs)(O.Z,{onClick:ev,children:[(0,n.jsx)(P.Z,{children:(0,n.jsx)(f.Z,{fontSize:"small"})}),"Delete"]})})]}),(0,n.jsx)(A.Z,{title:"Edit table view settings",placement:"bottom",children:(0,n.jsx)(S.Z,{color:"primary",onClick:()=>em(!0),startIcon:(0,n.jsx)(y.Z,{}),children:"Table Settings"})}),(null==x?void 0:null===(t=x.metadata)||void 0===t?void 0:t.redteam)&&(0,n.jsx)(A.Z,{title:"View vulnerability scan report",placement:"bottom",children:(0,n.jsx)(S.Z,{color:"primary",startIcon:(0,n.jsx)(w.Z,{}),onClick:()=>d.push("/report/?evalId=".concat(B)),children:"Vulnerability Report"})})]})})]})}),(0,n.jsx)(ex,{columnVisibility:eb}),(0,n.jsx)(eG,{maxTextLength:W,columnVisibility:eb,wordBreak:z,showStats:V,filterMode:Q,failureFilter:q,searchText:X,onFailureFilterToggle:K,onSearchTextChange:Y}),(0,n.jsx)(et,{open:eu,onClose:()=>ep(!1)}),(0,n.jsx)(eq,{open:en,onClose:()=>ea(!1),shareUrl:er}),(0,n.jsx)(eY,{open:eh,onClose:()=>em(!1)})]})}async function eQ(){let e=(0,h.createClientComponentClient)(),{data:{user:t}}=await e.auth.getUser();(0,g.Z)(t,"User not logged in");let{data:s,error:n}=await e.from("EvaluationResult").select("id, createdAt").eq("user_id",t.id).order("createdAt",{ascending:!1}).limit(100);return s||[]}async function e0(e){let t=(0,h.createClientComponentClient)(),{data:s,error:n}=await t.from("EvaluationResult").select("*").eq("id",e).single();return s}function e1(e){var t;let{fetchId:s,preloadedData:a,recentEvals:i,defaultEvalId:c}=e,h=(0,m.useRouter)(),{table:j,setTable:v,setConfig:f,setEvalId:b}=ee(),[y,Z]=l.useState(!1),[w,C]=l.useState(!1),[k,S]=l.useState(i||[]),N=async()=>{let e=await fetch("".concat(await (0,o.b)(),"/api/results"),{cache:"no-store"}),t=await e.json();return S(t.data),t.data},R=l.useCallback(async e=>{let t=await fetch("".concat(await (0,o.b)(),"/api/results/").concat(e),{cache:"no-store"}),s=await t.json();v(s.data.results.table),f(s.data.config),b(e)},[v,f,b]),I=async e=>{u.Ox?(Z(!1),h.push("/eval/remote:".concat(encodeURIComponent(e)))):h.push("/eval/?evalId=".concat(encodeURIComponent(e)))},[P,E]=l.useState(c||(null===(t=k[0])||void 0===t?void 0:t.id)),F=(0,m.useSearchParams)(),O=F?F.get("evalId"):null;return(l.useEffect(()=>{if(O){let e=async()=>{await R(O),Z(!0),E(O),N()};e()}else if(a){var e;v(null===(e=a.data.results)||void 0===e?void 0:e.table),f(a.data.config),Z(!0)}else if(s){let e=async()=>{var e;let t="".concat(r,"/api/eval/").concat(s);console.log("Fetching eval from remote server",t);let n=await fetch(t);if(!n.ok){C(!0);return}let l=await n.json();v(null===(e=l.data.results)||void 0===e?void 0:e.table),f(l.data.config),Z(!0)};e()}else if(u.T8)(0,o.b)().then(e=>{let t=(0,x.io)(e);return t.on("init",e=>{console.log("Initialized socket connection",e),Z(!0),v(null==e?void 0:e.results.table),f(null==e?void 0:e.config),N().then(e=>{var t,s;E(null===(t=e[0])||void 0===t?void 0:t.id),b(null===(s=e[0])||void 0===s?void 0:s.id)})}),t.on("update",e=>{console.log("Received data update",e),v(e.results.table),f(e.config),N().then(e=>{var t;let s=null===(t=e[0])||void 0===t?void 0:t.id;s&&(E(s),b(s))})}),()=>{t.disconnect()}});else if(u.Ox)eQ().then(e=>{S(e.map(e=>({id:e.id,label:e.createdAt}))),e.length>0&&e0(e[0].id).then(t=>{(0,g.Z)(t,"Eval not found");let s=t.results,n=t.config;E(e[0].id),v(s.table),f(n),Z(!0)})});else{let e=async()=>{let e=await N();if(!(e.length>0))return(0,n.jsx)("div",{className:"notice",children:"No evals yet. Share some evals to this server and they will appear here."});{let t=await (0,o.b)(),s=e[0].id,n=await fetch("".concat(t,"/api/results/").concat(s)),l=await n.json();v(l.data.results.table),f(l.data.config),Z(!0),E(s),b(s)}};e()}},[s,v,f,b,R,a,E,O]),w)?(0,n.jsx)("div",{className:"notice",children:"404 Eval not found"}):y&&j?(0,n.jsx)(d,{children:(0,n.jsx)(eK,{defaultEvalId:P,recentEvals:k,onRecentEvalSelected:I})}):(0,n.jsxs)("div",{className:"notice",children:[(0,n.jsx)("div",{children:(0,n.jsx)(p.Z,{size:22})}),(0,n.jsx)("div",{children:"Waiting for eval data"})]})}s(94455)},52428:function(e,t,s){"use strict";s.d(t,{Ox:function(){return a},T8:function(){return l},eA:function(){return r}});var n=s(77580);let l=!n.env.NEXT_PUBLIC_PROMPTFOO_BUILD_STANDALONE_SERVER,a=!!n.env.NEXT_PUBLIC_PROMPTFOO_USE_SUPABASE,r=""},47887:function(e,t,s){"use strict";s.d(t,{o:function(){return a}});var n=s(79685),l=s(74595);let a=(0,n.Ue)()((0,l.tJ)((e,t)=>({env:{},testCases:[],description:"",providers:[],prompts:[],defaultTest:{},evaluateOptions:{},setEnv:t=>e({env:t}),setTestCases:t=>e({testCases:t}),setDescription:t=>e({description:t}),setProviders:t=>e({providers:t}),setPrompts:t=>e({prompts:t}),setDefaultTest:t=>e({defaultTest:t}),setEvaluateOptions:t=>e({evaluateOptions:t}),setStateFromConfig:t=>{let s={};t.description&&(s.description=t.description||""),t.tests&&(s.testCases=t.tests),t.providers&&(s.providers=t.providers),t.prompts&&("string"==typeof t.prompts?s.prompts=[t.prompts]:Array.isArray(t.prompts)?s.prompts=t.prompts.filter(e=>"string"==typeof e&&!e.endsWith(".txt")&&!e.endsWith(".json")&&!e.endsWith(".yaml")):console.warn("Invalid prompts config",t.prompts)),t.defaultTest&&(s.defaultTest=t.defaultTest),t.evaluateOptions&&(s.evaluateOptions=t.evaluateOptions),e(s)},getTestSuite:()=>{let{description:e,testCases:s,providers:n,prompts:l,env:a}=t();return{env:a,description:e,providers:n,prompts:l,tests:s}}}),{name:"promptfoo",skipHydration:!0}))},55974:function(){},94455:function(){},16658:function(){},93207:function(){},58022:function(){}}]);
@@ -1 +1 @@
1
- (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[716],{11520:function(t,e,r){"use strict";r.d(e,{Z:function(){return S}});var a=r(94312),n=r(79042),i=r(14978),o=r(53468),s=r(46213),u=r(3982),l=r(225),d=r(48976),p=r(14059),h=r(23437),c=r(24004);let m=["className","component","disableGutters","fixed","maxWidth","classes"],g=(0,h.Z)(),f=(0,p.Z)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,e[`maxWidth${(0,l.Z)(String(r.maxWidth))}`],r.fixed&&e.fixed,r.disableGutters&&e.disableGutters]}}),x=t=>(0,d.Z)({props:t,name:"MuiContainer",defaultTheme:g}),b=(t,e)=>{let{classes:r,fixed:a,disableGutters:n,maxWidth:i}=t,o={root:["root",i&&`maxWidth${(0,l.Z)(String(i))}`,a&&"fixed",n&&"disableGutters"]};return(0,u.Z)(o,t=>(0,s.ZP)(e,t),r)};var v=r(81427),y=r(31608),Z=r(89020);let W=function(t={}){let{createStyledComponent:e=f,useThemeProps:r=x,componentName:s="MuiContainer"}=t,u=e(({theme:t,ownerState:e})=>(0,n.Z)({width:"100%",marginLeft:"auto",boxSizing:"border-box",marginRight:"auto",display:"block"},!e.disableGutters&&{paddingLeft:t.spacing(2),paddingRight:t.spacing(2),[t.breakpoints.up("sm")]:{paddingLeft:t.spacing(3),paddingRight:t.spacing(3)}}),({theme:t,ownerState:e})=>e.fixed&&Object.keys(t.breakpoints.values).reduce((e,r)=>{let a=t.breakpoints.values[r];return 0!==a&&(e[t.breakpoints.up(r)]={maxWidth:`${a}${t.breakpoints.unit}`}),e},{}),({theme:t,ownerState:e})=>(0,n.Z)({},"xs"===e.maxWidth&&{[t.breakpoints.up("xs")]:{maxWidth:Math.max(t.breakpoints.values.xs,444)}},e.maxWidth&&"xs"!==e.maxWidth&&{[t.breakpoints.up(e.maxWidth)]:{maxWidth:`${t.breakpoints.values[e.maxWidth]}${t.breakpoints.unit}`}})),l=i.forwardRef(function(t,e){let i=r(t),{className:l,component:d="div",disableGutters:p=!1,fixed:h=!1,maxWidth:g="lg"}=i,f=(0,a.Z)(i,m),x=(0,n.Z)({},i,{component:d,disableGutters:p,fixed:h,maxWidth:g}),v=b(x,s);return(0,c.jsx)(u,(0,n.Z)({as:d,ownerState:x,className:(0,o.Z)(v.root,l),ref:e},f))});return l}({createStyledComponent:(0,y.ZP)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,e[`maxWidth${(0,v.Z)(String(r.maxWidth))}`],r.fixed&&e.fixed,r.disableGutters&&e.disableGutters]}}),useThemeProps:t=>(0,Z.Z)({props:t,name:"MuiContainer"})});var S=W},63147:function(t,e,r){"use strict";r.d(e,{Z:function(){return S}});var a=r(94312),n=r(79042),i=r(14978),o=r(53468),s=r(86251),u=r(3982),l=r(31608),d=r(89020),p=r(81427),h=r(25800),c=r(46213);function m(t){return(0,c.ZP)("MuiTypography",t)}(0,h.Z)("MuiTypography",["root","h1","h2","h3","h4","h5","h6","subtitle1","subtitle2","body1","body2","inherit","button","caption","overline","alignLeft","alignRight","alignCenter","alignJustify","noWrap","gutterBottom","paragraph"]);var g=r(24004);let f=["align","className","component","gutterBottom","noWrap","paragraph","variant","variantMapping"],x=t=>{let{align:e,gutterBottom:r,noWrap:a,paragraph:n,variant:i,classes:o}=t,s={root:["root",i,"inherit"!==t.align&&`align${(0,p.Z)(e)}`,r&&"gutterBottom",a&&"noWrap",n&&"paragraph"]};return(0,u.Z)(s,m,o)},b=(0,l.ZP)("span",{name:"MuiTypography",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,r.variant&&e[r.variant],"inherit"!==r.align&&e[`align${(0,p.Z)(r.align)}`],r.noWrap&&e.noWrap,r.gutterBottom&&e.gutterBottom,r.paragraph&&e.paragraph]}})(({theme:t,ownerState:e})=>(0,n.Z)({margin:0},"inherit"===e.variant&&{font:"inherit"},"inherit"!==e.variant&&t.typography[e.variant],"inherit"!==e.align&&{textAlign:e.align},e.noWrap&&{overflow:"hidden",textOverflow:"ellipsis",whiteSpace:"nowrap"},e.gutterBottom&&{marginBottom:"0.35em"},e.paragraph&&{marginBottom:16})),v={h1:"h1",h2:"h2",h3:"h3",h4:"h4",h5:"h5",h6:"h6",subtitle1:"h6",subtitle2:"h6",body1:"p",body2:"p",inherit:"p"},y={primary:"primary.main",textPrimary:"text.primary",secondary:"secondary.main",textSecondary:"text.secondary",error:"error.main"},Z=t=>y[t]||t,W=i.forwardRef(function(t,e){let r=(0,d.Z)({props:t,name:"MuiTypography"}),i=Z(r.color),u=(0,s.Z)((0,n.Z)({},r,{color:i})),{align:l="inherit",className:p,component:h,gutterBottom:c=!1,noWrap:m=!1,paragraph:y=!1,variant:W="body1",variantMapping:S=v}=u,C=(0,a.Z)(u,f),w=(0,n.Z)({},u,{align:l,color:i,className:p,component:h,gutterBottom:c,noWrap:m,paragraph:y,variant:W,variantMapping:S}),k=h||(y?"p":S[W]||v[W])||"span",R=x(w);return(0,g.jsx)(b,(0,n.Z)({as:k,ref:e,ownerState:w,className:(0,o.Z)(R.root,p)},C))});var S=W},40462:function(t,e,r){Promise.resolve().then(r.bind(r,92030))},92030:function(t,e,r){"use strict";r.r(e),r.d(e,{default:function(){return c}});var a=r(24004),n=r(14978),i=r(38640),o=r.n(i),s=r(40982),u=r(67339),l=r(11520),d=r(63147),p=r(28891),h=r(1445);function c(){let t=(0,p.useRouter)(),[e,r]=n.useState(""),[i,c]=n.useState(""),[m,g]=n.useState(""),[f,x]=n.useState(!1),[b,v]=n.useState(null),y=n.useCallback(async()=>{let{data:t,error:e}=await h.OQ.auth.refreshSession();t&&v(t.user)},[]);n.useEffect(()=>{y()},[y]);let Z=async r=>{if(x(!0),g(""),r.stopPropagation(),r.preventDefault(),!b){let{data:r,error:a}=await h.OQ.auth.signInWithPassword({email:e,password:i});a?g(a.message):t.push("/")}x(!1)};return b?(0,a.jsxs)("div",{style:{textAlign:"center"},children:["You're already logged in as ",b.email,"."]}):(0,a.jsxs)(l.Z,{component:"main",maxWidth:"xs",children:[(0,a.jsx)(d.Z,{component:"h1",variant:"h5",children:"Sign in"}),(0,a.jsx)("form",{children:!b&&(0,a.jsxs)(a.Fragment,{children:[(0,a.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,id:"email",label:"Email Address",name:"email",autoComplete:"email",autoFocus:!0,onChange:t=>r(t.target.value),value:e,error:!!m}),(0,a.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,name:"password",label:"Password",type:"password",id:"password",autoComplete:"current-password",onChange:t=>c(t.target.value),value:i,error:!!m,helperText:m}),(0,a.jsx)(s.Z,{type:"submit",disabled:f,sx:{marginTop:"1em"},fullWidth:!0,variant:"contained",color:"primary",onClick:Z,children:"Sign In"}),(0,a.jsxs)("p",{children:["Don't have an account yet? ",(0,a.jsx)(o(),{href:"/auth/signup",children:"Sign up"})]})]})})]})}},1445:function(t,e,r){"use strict";r.d(e,{Ho:function(){return p},OQ:function(){return o},aC:function(){return u}});var a=r(24004),n=r(14978),i=r(12594);let o=(0,i.createClientComponentClient)(),s=(0,n.createContext)({}),u=()=>(0,n.useContext)(s),l=(t,e)=>o.auth.signInWithPassword({email:t,password:e}),d=()=>o.auth.signOut(),p=t=>{let{children:e}=t,[r,i]=(0,n.useState)(null),[u,p]=(0,n.useState)(!1),h=(0,n.useCallback)(async()=>{let{data:t,error:e}=await o.auth.refreshSession();t&&i(t.user)},[]);return(0,n.useEffect)(()=>{h();let{data:t}=o.auth.onAuthStateChange(async(t,e)=>{"PASSWORD_RECOVERY"==t?p(!1):"SIGNED_IN"===t&&e?(i(e.user),p(!0)):"SIGNED_OUT"===t&&(p(!1),i(null))});return()=>{t.subscription.unsubscribe()}},[h]),(0,a.jsx)(s.Provider,{value:{loggedIn:u,user:r,login:l,logout:d},children:e})}},28891:function(t,e,r){t.exports=r(98556)}},function(t){t.O(0,[540,640,215,620,304,339,954,470,730,744],function(){return t(t.s=40462)}),_N_E=t.O()}]);
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[716],{11520:function(t,e,r){"use strict";r.d(e,{Z:function(){return S}});var a=r(94312),n=r(79042),i=r(14978),o=r(53468),s=r(46213),u=r(3982),l=r(225),d=r(48976),p=r(14059),h=r(23437),c=r(24004);let m=["className","component","disableGutters","fixed","maxWidth","classes"],g=(0,h.Z)(),f=(0,p.Z)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,e[`maxWidth${(0,l.Z)(String(r.maxWidth))}`],r.fixed&&e.fixed,r.disableGutters&&e.disableGutters]}}),x=t=>(0,d.Z)({props:t,name:"MuiContainer",defaultTheme:g}),b=(t,e)=>{let{classes:r,fixed:a,disableGutters:n,maxWidth:i}=t,o={root:["root",i&&`maxWidth${(0,l.Z)(String(i))}`,a&&"fixed",n&&"disableGutters"]};return(0,u.Z)(o,t=>(0,s.ZP)(e,t),r)};var v=r(81427),y=r(31608),Z=r(89020);let W=function(t={}){let{createStyledComponent:e=f,useThemeProps:r=x,componentName:s="MuiContainer"}=t,u=e(({theme:t,ownerState:e})=>(0,n.Z)({width:"100%",marginLeft:"auto",boxSizing:"border-box",marginRight:"auto",display:"block"},!e.disableGutters&&{paddingLeft:t.spacing(2),paddingRight:t.spacing(2),[t.breakpoints.up("sm")]:{paddingLeft:t.spacing(3),paddingRight:t.spacing(3)}}),({theme:t,ownerState:e})=>e.fixed&&Object.keys(t.breakpoints.values).reduce((e,r)=>{let a=t.breakpoints.values[r];return 0!==a&&(e[t.breakpoints.up(r)]={maxWidth:`${a}${t.breakpoints.unit}`}),e},{}),({theme:t,ownerState:e})=>(0,n.Z)({},"xs"===e.maxWidth&&{[t.breakpoints.up("xs")]:{maxWidth:Math.max(t.breakpoints.values.xs,444)}},e.maxWidth&&"xs"!==e.maxWidth&&{[t.breakpoints.up(e.maxWidth)]:{maxWidth:`${t.breakpoints.values[e.maxWidth]}${t.breakpoints.unit}`}})),l=i.forwardRef(function(t,e){let i=r(t),{className:l,component:d="div",disableGutters:p=!1,fixed:h=!1,maxWidth:g="lg"}=i,f=(0,a.Z)(i,m),x=(0,n.Z)({},i,{component:d,disableGutters:p,fixed:h,maxWidth:g}),v=b(x,s);return(0,c.jsx)(u,(0,n.Z)({as:d,ownerState:x,className:(0,o.Z)(v.root,l),ref:e},f))});return l}({createStyledComponent:(0,y.ZP)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,e[`maxWidth${(0,v.Z)(String(r.maxWidth))}`],r.fixed&&e.fixed,r.disableGutters&&e.disableGutters]}}),useThemeProps:t=>(0,Z.Z)({props:t,name:"MuiContainer"})});var S=W},63147:function(t,e,r){"use strict";r.d(e,{Z:function(){return S}});var a=r(94312),n=r(79042),i=r(14978),o=r(53468),s=r(86251),u=r(3982),l=r(31608),d=r(89020),p=r(81427),h=r(25800),c=r(46213);function m(t){return(0,c.ZP)("MuiTypography",t)}(0,h.Z)("MuiTypography",["root","h1","h2","h3","h4","h5","h6","subtitle1","subtitle2","body1","body2","inherit","button","caption","overline","alignLeft","alignRight","alignCenter","alignJustify","noWrap","gutterBottom","paragraph"]);var g=r(24004);let f=["align","className","component","gutterBottom","noWrap","paragraph","variant","variantMapping"],x=t=>{let{align:e,gutterBottom:r,noWrap:a,paragraph:n,variant:i,classes:o}=t,s={root:["root",i,"inherit"!==t.align&&`align${(0,p.Z)(e)}`,r&&"gutterBottom",a&&"noWrap",n&&"paragraph"]};return(0,u.Z)(s,m,o)},b=(0,l.ZP)("span",{name:"MuiTypography",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:r}=t;return[e.root,r.variant&&e[r.variant],"inherit"!==r.align&&e[`align${(0,p.Z)(r.align)}`],r.noWrap&&e.noWrap,r.gutterBottom&&e.gutterBottom,r.paragraph&&e.paragraph]}})(({theme:t,ownerState:e})=>(0,n.Z)({margin:0},"inherit"===e.variant&&{font:"inherit"},"inherit"!==e.variant&&t.typography[e.variant],"inherit"!==e.align&&{textAlign:e.align},e.noWrap&&{overflow:"hidden",textOverflow:"ellipsis",whiteSpace:"nowrap"},e.gutterBottom&&{marginBottom:"0.35em"},e.paragraph&&{marginBottom:16})),v={h1:"h1",h2:"h2",h3:"h3",h4:"h4",h5:"h5",h6:"h6",subtitle1:"h6",subtitle2:"h6",body1:"p",body2:"p",inherit:"p"},y={primary:"primary.main",textPrimary:"text.primary",secondary:"secondary.main",textSecondary:"text.secondary",error:"error.main"},Z=t=>y[t]||t,W=i.forwardRef(function(t,e){let r=(0,d.Z)({props:t,name:"MuiTypography"}),i=Z(r.color),u=(0,s.Z)((0,n.Z)({},r,{color:i})),{align:l="inherit",className:p,component:h,gutterBottom:c=!1,noWrap:m=!1,paragraph:y=!1,variant:W="body1",variantMapping:S=v}=u,C=(0,a.Z)(u,f),w=(0,n.Z)({},u,{align:l,color:i,className:p,component:h,gutterBottom:c,noWrap:m,paragraph:y,variant:W,variantMapping:S}),k=h||(y?"p":S[W]||v[W])||"span",R=x(w);return(0,g.jsx)(b,(0,n.Z)({as:k,ref:e,ownerState:w,className:(0,o.Z)(R.root,p)},C))});var S=W},40462:function(t,e,r){Promise.resolve().then(r.bind(r,92030))},92030:function(t,e,r){"use strict";r.r(e),r.d(e,{default:function(){return c}});var a=r(24004),n=r(14978),i=r(1445),o=r(40982),s=r(11520),u=r(67339),l=r(63147),d=r(38640),p=r.n(d),h=r(28891);function c(){let t=(0,h.useRouter)(),[e,r]=n.useState(""),[d,c]=n.useState(""),[m,g]=n.useState(""),[f,x]=n.useState(!1),[b,v]=n.useState(null),y=n.useCallback(async()=>{let{data:t,error:e}=await i.OQ.auth.refreshSession();t&&v(t.user)},[]);n.useEffect(()=>{y()},[y]);let Z=async r=>{if(x(!0),g(""),r.stopPropagation(),r.preventDefault(),!b){let{data:r,error:a}=await i.OQ.auth.signInWithPassword({email:e,password:d});a?g(a.message):t.push("/")}x(!1)};return b?(0,a.jsxs)("div",{style:{textAlign:"center"},children:["You're already logged in as ",b.email,"."]}):(0,a.jsxs)(s.Z,{component:"main",maxWidth:"xs",children:[(0,a.jsx)(l.Z,{component:"h1",variant:"h5",children:"Sign in"}),(0,a.jsx)("form",{children:!b&&(0,a.jsxs)(a.Fragment,{children:[(0,a.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,id:"email",label:"Email Address",name:"email",autoComplete:"email",autoFocus:!0,onChange:t=>r(t.target.value),value:e,error:!!m}),(0,a.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,name:"password",label:"Password",type:"password",id:"password",autoComplete:"current-password",onChange:t=>c(t.target.value),value:d,error:!!m,helperText:m}),(0,a.jsx)(o.Z,{type:"submit",disabled:f,sx:{marginTop:"1em"},fullWidth:!0,variant:"contained",color:"primary",onClick:Z,children:"Sign In"}),(0,a.jsxs)("p",{children:["Don't have an account yet? ",(0,a.jsx)(p(),{href:"/auth/signup",children:"Sign up"})]})]})})]})}},1445:function(t,e,r){"use strict";r.d(e,{Ho:function(){return p},OQ:function(){return o},aC:function(){return u}});var a=r(24004),n=r(14978),i=r(12594);let o=(0,i.createClientComponentClient)(),s=(0,n.createContext)({}),u=()=>(0,n.useContext)(s),l=(t,e)=>o.auth.signInWithPassword({email:t,password:e}),d=()=>o.auth.signOut(),p=t=>{let{children:e}=t,[r,i]=(0,n.useState)(null),[u,p]=(0,n.useState)(!1),h=(0,n.useCallback)(async()=>{let{data:t,error:e}=await o.auth.refreshSession();t&&i(t.user)},[]);return(0,n.useEffect)(()=>{h();let{data:t}=o.auth.onAuthStateChange(async(t,e)=>{"PASSWORD_RECOVERY"==t?p(!1):"SIGNED_IN"===t&&e?(i(e.user),p(!0)):"SIGNED_OUT"===t&&(p(!1),i(null))});return()=>{t.subscription.unsubscribe()}},[h]),(0,a.jsx)(s.Provider,{value:{loggedIn:u,user:r,login:l,logout:d},children:e})}},28891:function(t,e,r){t.exports=r(98556)}},function(t){t.O(0,[540,640,215,620,304,339,954,470,730,744],function(){return t(t.s=40462)}),_N_E=t.O()}]);
@@ -1 +1 @@
1
- (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[271],{11520:function(t,e,a){"use strict";a.d(e,{Z:function(){return W}});var r=a(94312),n=a(79042),i=a(14978),o=a(53468),s=a(46213),u=a(3982),l=a(225),d=a(48976),p=a(14059),h=a(23437),c=a(24004);let m=["className","component","disableGutters","fixed","maxWidth","classes"],g=(0,h.Z)(),f=(0,p.Z)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,e[`maxWidth${(0,l.Z)(String(a.maxWidth))}`],a.fixed&&e.fixed,a.disableGutters&&e.disableGutters]}}),x=t=>(0,d.Z)({props:t,name:"MuiContainer",defaultTheme:g}),b=(t,e)=>{let{classes:a,fixed:r,disableGutters:n,maxWidth:i}=t,o={root:["root",i&&`maxWidth${(0,l.Z)(String(i))}`,r&&"fixed",n&&"disableGutters"]};return(0,u.Z)(o,t=>(0,s.ZP)(e,t),a)};var v=a(81427),y=a(31608),Z=a(89020);let S=function(t={}){let{createStyledComponent:e=f,useThemeProps:a=x,componentName:s="MuiContainer"}=t,u=e(({theme:t,ownerState:e})=>(0,n.Z)({width:"100%",marginLeft:"auto",boxSizing:"border-box",marginRight:"auto",display:"block"},!e.disableGutters&&{paddingLeft:t.spacing(2),paddingRight:t.spacing(2),[t.breakpoints.up("sm")]:{paddingLeft:t.spacing(3),paddingRight:t.spacing(3)}}),({theme:t,ownerState:e})=>e.fixed&&Object.keys(t.breakpoints.values).reduce((e,a)=>{let r=t.breakpoints.values[a];return 0!==r&&(e[t.breakpoints.up(a)]={maxWidth:`${r}${t.breakpoints.unit}`}),e},{}),({theme:t,ownerState:e})=>(0,n.Z)({},"xs"===e.maxWidth&&{[t.breakpoints.up("xs")]:{maxWidth:Math.max(t.breakpoints.values.xs,444)}},e.maxWidth&&"xs"!==e.maxWidth&&{[t.breakpoints.up(e.maxWidth)]:{maxWidth:`${t.breakpoints.values[e.maxWidth]}${t.breakpoints.unit}`}})),l=i.forwardRef(function(t,e){let i=a(t),{className:l,component:d="div",disableGutters:p=!1,fixed:h=!1,maxWidth:g="lg"}=i,f=(0,r.Z)(i,m),x=(0,n.Z)({},i,{component:d,disableGutters:p,fixed:h,maxWidth:g}),v=b(x,s);return(0,c.jsx)(u,(0,n.Z)({as:d,ownerState:x,className:(0,o.Z)(v.root,l),ref:e},f))});return l}({createStyledComponent:(0,y.ZP)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,e[`maxWidth${(0,v.Z)(String(a.maxWidth))}`],a.fixed&&e.fixed,a.disableGutters&&e.disableGutters]}}),useThemeProps:t=>(0,Z.Z)({props:t,name:"MuiContainer"})});var W=S},63147:function(t,e,a){"use strict";a.d(e,{Z:function(){return W}});var r=a(94312),n=a(79042),i=a(14978),o=a(53468),s=a(86251),u=a(3982),l=a(31608),d=a(89020),p=a(81427),h=a(25800),c=a(46213);function m(t){return(0,c.ZP)("MuiTypography",t)}(0,h.Z)("MuiTypography",["root","h1","h2","h3","h4","h5","h6","subtitle1","subtitle2","body1","body2","inherit","button","caption","overline","alignLeft","alignRight","alignCenter","alignJustify","noWrap","gutterBottom","paragraph"]);var g=a(24004);let f=["align","className","component","gutterBottom","noWrap","paragraph","variant","variantMapping"],x=t=>{let{align:e,gutterBottom:a,noWrap:r,paragraph:n,variant:i,classes:o}=t,s={root:["root",i,"inherit"!==t.align&&`align${(0,p.Z)(e)}`,a&&"gutterBottom",r&&"noWrap",n&&"paragraph"]};return(0,u.Z)(s,m,o)},b=(0,l.ZP)("span",{name:"MuiTypography",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,a.variant&&e[a.variant],"inherit"!==a.align&&e[`align${(0,p.Z)(a.align)}`],a.noWrap&&e.noWrap,a.gutterBottom&&e.gutterBottom,a.paragraph&&e.paragraph]}})(({theme:t,ownerState:e})=>(0,n.Z)({margin:0},"inherit"===e.variant&&{font:"inherit"},"inherit"!==e.variant&&t.typography[e.variant],"inherit"!==e.align&&{textAlign:e.align},e.noWrap&&{overflow:"hidden",textOverflow:"ellipsis",whiteSpace:"nowrap"},e.gutterBottom&&{marginBottom:"0.35em"},e.paragraph&&{marginBottom:16})),v={h1:"h1",h2:"h2",h3:"h3",h4:"h4",h5:"h5",h6:"h6",subtitle1:"h6",subtitle2:"h6",body1:"p",body2:"p",inherit:"p"},y={primary:"primary.main",textPrimary:"text.primary",secondary:"secondary.main",textSecondary:"text.secondary",error:"error.main"},Z=t=>y[t]||t,S=i.forwardRef(function(t,e){let a=(0,d.Z)({props:t,name:"MuiTypography"}),i=Z(a.color),u=(0,s.Z)((0,n.Z)({},a,{color:i})),{align:l="inherit",className:p,component:h,gutterBottom:c=!1,noWrap:m=!1,paragraph:y=!1,variant:S="body1",variantMapping:W=v}=u,C=(0,r.Z)(u,f),w=(0,n.Z)({},u,{align:l,color:i,className:p,component:h,gutterBottom:c,noWrap:m,paragraph:y,variant:S,variantMapping:W}),k=h||(y?"p":W[S]||v[S])||"span",R=x(w);return(0,g.jsx)(b,(0,n.Z)({as:k,ref:e,ownerState:w,className:(0,o.Z)(R.root,p)},C))});var W=S},38874:function(t,e,a){Promise.resolve().then(a.bind(a,78567))},78567:function(t,e,a){"use strict";a.r(e),a.d(e,{default:function(){return c}});var r=a(24004),n=a(14978),i=a(38640),o=a.n(i),s=a(40982),u=a(67339),l=a(11520),d=a(63147),p=a(28891),h=a(1445);function c(){let t=(0,p.useRouter)(),[e,a]=n.useState(""),[i,c]=n.useState(""),[m,g]=n.useState(""),[f,x]=n.useState(!1),[b,v]=n.useState(null),y=n.useCallback(async()=>{let{data:t,error:e}=await h.OQ.auth.refreshSession();t&&v(t.user)},[]);n.useEffect(()=>{y()},[y]);let Z=async a=>{if(x(!0),g(""),a.stopPropagation(),a.preventDefault(),!b){let{data:a,error:r}=await h.OQ.auth.signUp({email:e,password:i,options:{emailRedirectTo:"".concat(location.origin,"/auth/callback")}});r?g(r.message):t.push("/")}x(!1)};return(0,r.jsxs)(l.Z,{component:"main",maxWidth:"xs",children:[(0,r.jsx)(d.Z,{component:"h1",variant:"h5",children:"Sign up"}),(0,r.jsx)("p",{children:(0,r.jsx)(d.Z,{variant:"body1",children:"An account helps you easily save, view, and share your evals."})}),(0,r.jsxs)("form",{children:[!b&&(0,r.jsxs)(r.Fragment,{children:[(0,r.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,id:"email",label:"Email Address",name:"email",autoComplete:"email",autoFocus:!0,onChange:t=>a(t.target.value),value:e,error:!!m}),(0,r.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,name:"password",label:"Password",type:"password",id:"password",autoComplete:"current-password",onChange:t=>c(t.target.value),value:i,error:!!m,helperText:m}),(0,r.jsx)(s.Z,{type:"submit",disabled:f,sx:{marginTop:"1em"},fullWidth:!0,variant:"contained",color:"primary",onClick:Z,children:"Sign Up"})]}),(0,r.jsxs)("p",{children:["Already have an account? ",(0,r.jsx)(o(),{href:"/auth/login",children:"Sign in"})]})]})]})}},1445:function(t,e,a){"use strict";a.d(e,{Ho:function(){return p},OQ:function(){return o},aC:function(){return u}});var r=a(24004),n=a(14978),i=a(12594);let o=(0,i.createClientComponentClient)(),s=(0,n.createContext)({}),u=()=>(0,n.useContext)(s),l=(t,e)=>o.auth.signInWithPassword({email:t,password:e}),d=()=>o.auth.signOut(),p=t=>{let{children:e}=t,[a,i]=(0,n.useState)(null),[u,p]=(0,n.useState)(!1),h=(0,n.useCallback)(async()=>{let{data:t,error:e}=await o.auth.refreshSession();t&&i(t.user)},[]);return(0,n.useEffect)(()=>{h();let{data:t}=o.auth.onAuthStateChange(async(t,e)=>{"PASSWORD_RECOVERY"==t?p(!1):"SIGNED_IN"===t&&e?(i(e.user),p(!0)):"SIGNED_OUT"===t&&(p(!1),i(null))});return()=>{t.subscription.unsubscribe()}},[h]),(0,r.jsx)(s.Provider,{value:{loggedIn:u,user:a,login:l,logout:d},children:e})}},28891:function(t,e,a){t.exports=a(98556)}},function(t){t.O(0,[540,640,215,620,304,339,954,470,730,744],function(){return t(t.s=38874)}),_N_E=t.O()}]);
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[271],{11520:function(t,e,a){"use strict";a.d(e,{Z:function(){return W}});var r=a(94312),n=a(79042),i=a(14978),o=a(53468),s=a(46213),u=a(3982),l=a(225),d=a(48976),p=a(14059),h=a(23437),c=a(24004);let m=["className","component","disableGutters","fixed","maxWidth","classes"],g=(0,h.Z)(),f=(0,p.Z)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,e[`maxWidth${(0,l.Z)(String(a.maxWidth))}`],a.fixed&&e.fixed,a.disableGutters&&e.disableGutters]}}),x=t=>(0,d.Z)({props:t,name:"MuiContainer",defaultTheme:g}),b=(t,e)=>{let{classes:a,fixed:r,disableGutters:n,maxWidth:i}=t,o={root:["root",i&&`maxWidth${(0,l.Z)(String(i))}`,r&&"fixed",n&&"disableGutters"]};return(0,u.Z)(o,t=>(0,s.ZP)(e,t),a)};var v=a(81427),y=a(31608),Z=a(89020);let S=function(t={}){let{createStyledComponent:e=f,useThemeProps:a=x,componentName:s="MuiContainer"}=t,u=e(({theme:t,ownerState:e})=>(0,n.Z)({width:"100%",marginLeft:"auto",boxSizing:"border-box",marginRight:"auto",display:"block"},!e.disableGutters&&{paddingLeft:t.spacing(2),paddingRight:t.spacing(2),[t.breakpoints.up("sm")]:{paddingLeft:t.spacing(3),paddingRight:t.spacing(3)}}),({theme:t,ownerState:e})=>e.fixed&&Object.keys(t.breakpoints.values).reduce((e,a)=>{let r=t.breakpoints.values[a];return 0!==r&&(e[t.breakpoints.up(a)]={maxWidth:`${r}${t.breakpoints.unit}`}),e},{}),({theme:t,ownerState:e})=>(0,n.Z)({},"xs"===e.maxWidth&&{[t.breakpoints.up("xs")]:{maxWidth:Math.max(t.breakpoints.values.xs,444)}},e.maxWidth&&"xs"!==e.maxWidth&&{[t.breakpoints.up(e.maxWidth)]:{maxWidth:`${t.breakpoints.values[e.maxWidth]}${t.breakpoints.unit}`}})),l=i.forwardRef(function(t,e){let i=a(t),{className:l,component:d="div",disableGutters:p=!1,fixed:h=!1,maxWidth:g="lg"}=i,f=(0,r.Z)(i,m),x=(0,n.Z)({},i,{component:d,disableGutters:p,fixed:h,maxWidth:g}),v=b(x,s);return(0,c.jsx)(u,(0,n.Z)({as:d,ownerState:x,className:(0,o.Z)(v.root,l),ref:e},f))});return l}({createStyledComponent:(0,y.ZP)("div",{name:"MuiContainer",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,e[`maxWidth${(0,v.Z)(String(a.maxWidth))}`],a.fixed&&e.fixed,a.disableGutters&&e.disableGutters]}}),useThemeProps:t=>(0,Z.Z)({props:t,name:"MuiContainer"})});var W=S},63147:function(t,e,a){"use strict";a.d(e,{Z:function(){return W}});var r=a(94312),n=a(79042),i=a(14978),o=a(53468),s=a(86251),u=a(3982),l=a(31608),d=a(89020),p=a(81427),h=a(25800),c=a(46213);function m(t){return(0,c.ZP)("MuiTypography",t)}(0,h.Z)("MuiTypography",["root","h1","h2","h3","h4","h5","h6","subtitle1","subtitle2","body1","body2","inherit","button","caption","overline","alignLeft","alignRight","alignCenter","alignJustify","noWrap","gutterBottom","paragraph"]);var g=a(24004);let f=["align","className","component","gutterBottom","noWrap","paragraph","variant","variantMapping"],x=t=>{let{align:e,gutterBottom:a,noWrap:r,paragraph:n,variant:i,classes:o}=t,s={root:["root",i,"inherit"!==t.align&&`align${(0,p.Z)(e)}`,a&&"gutterBottom",r&&"noWrap",n&&"paragraph"]};return(0,u.Z)(s,m,o)},b=(0,l.ZP)("span",{name:"MuiTypography",slot:"Root",overridesResolver:(t,e)=>{let{ownerState:a}=t;return[e.root,a.variant&&e[a.variant],"inherit"!==a.align&&e[`align${(0,p.Z)(a.align)}`],a.noWrap&&e.noWrap,a.gutterBottom&&e.gutterBottom,a.paragraph&&e.paragraph]}})(({theme:t,ownerState:e})=>(0,n.Z)({margin:0},"inherit"===e.variant&&{font:"inherit"},"inherit"!==e.variant&&t.typography[e.variant],"inherit"!==e.align&&{textAlign:e.align},e.noWrap&&{overflow:"hidden",textOverflow:"ellipsis",whiteSpace:"nowrap"},e.gutterBottom&&{marginBottom:"0.35em"},e.paragraph&&{marginBottom:16})),v={h1:"h1",h2:"h2",h3:"h3",h4:"h4",h5:"h5",h6:"h6",subtitle1:"h6",subtitle2:"h6",body1:"p",body2:"p",inherit:"p"},y={primary:"primary.main",textPrimary:"text.primary",secondary:"secondary.main",textSecondary:"text.secondary",error:"error.main"},Z=t=>y[t]||t,S=i.forwardRef(function(t,e){let a=(0,d.Z)({props:t,name:"MuiTypography"}),i=Z(a.color),u=(0,s.Z)((0,n.Z)({},a,{color:i})),{align:l="inherit",className:p,component:h,gutterBottom:c=!1,noWrap:m=!1,paragraph:y=!1,variant:S="body1",variantMapping:W=v}=u,C=(0,r.Z)(u,f),w=(0,n.Z)({},u,{align:l,color:i,className:p,component:h,gutterBottom:c,noWrap:m,paragraph:y,variant:S,variantMapping:W}),k=h||(y?"p":W[S]||v[S])||"span",R=x(w);return(0,g.jsx)(b,(0,n.Z)({as:k,ref:e,ownerState:w,className:(0,o.Z)(R.root,p)},C))});var W=S},38874:function(t,e,a){Promise.resolve().then(a.bind(a,78567))},78567:function(t,e,a){"use strict";a.r(e),a.d(e,{default:function(){return c}});var r=a(24004),n=a(14978),i=a(1445),o=a(40982),s=a(11520),u=a(67339),l=a(63147),d=a(38640),p=a.n(d),h=a(28891);function c(){let t=(0,h.useRouter)(),[e,a]=n.useState(""),[d,c]=n.useState(""),[m,g]=n.useState(""),[f,x]=n.useState(!1),[b,v]=n.useState(null),y=n.useCallback(async()=>{let{data:t,error:e}=await i.OQ.auth.refreshSession();t&&v(t.user)},[]);n.useEffect(()=>{y()},[y]);let Z=async a=>{if(x(!0),g(""),a.stopPropagation(),a.preventDefault(),!b){let{data:a,error:r}=await i.OQ.auth.signUp({email:e,password:d,options:{emailRedirectTo:"".concat(location.origin,"/auth/callback")}});r?g(r.message):t.push("/")}x(!1)};return(0,r.jsxs)(s.Z,{component:"main",maxWidth:"xs",children:[(0,r.jsx)(l.Z,{component:"h1",variant:"h5",children:"Sign up"}),(0,r.jsx)("p",{children:(0,r.jsx)(l.Z,{variant:"body1",children:"An account helps you easily save, view, and share your evals."})}),(0,r.jsxs)("form",{children:[!b&&(0,r.jsxs)(r.Fragment,{children:[(0,r.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,id:"email",label:"Email Address",name:"email",autoComplete:"email",autoFocus:!0,onChange:t=>a(t.target.value),value:e,error:!!m}),(0,r.jsx)(u.Z,{disabled:f,variant:"outlined",margin:"normal",required:!0,fullWidth:!0,name:"password",label:"Password",type:"password",id:"password",autoComplete:"current-password",onChange:t=>c(t.target.value),value:d,error:!!m,helperText:m}),(0,r.jsx)(o.Z,{type:"submit",disabled:f,sx:{marginTop:"1em"},fullWidth:!0,variant:"contained",color:"primary",onClick:Z,children:"Sign Up"})]}),(0,r.jsxs)("p",{children:["Already have an account? ",(0,r.jsx)(p(),{href:"/auth/login",children:"Sign in"})]})]})]})}},1445:function(t,e,a){"use strict";a.d(e,{Ho:function(){return p},OQ:function(){return o},aC:function(){return u}});var r=a(24004),n=a(14978),i=a(12594);let o=(0,i.createClientComponentClient)(),s=(0,n.createContext)({}),u=()=>(0,n.useContext)(s),l=(t,e)=>o.auth.signInWithPassword({email:t,password:e}),d=()=>o.auth.signOut(),p=t=>{let{children:e}=t,[a,i]=(0,n.useState)(null),[u,p]=(0,n.useState)(!1),h=(0,n.useCallback)(async()=>{let{data:t,error:e}=await o.auth.refreshSession();t&&i(t.user)},[]);return(0,n.useEffect)(()=>{h();let{data:t}=o.auth.onAuthStateChange(async(t,e)=>{"PASSWORD_RECOVERY"==t?p(!1):"SIGNED_IN"===t&&e?(i(e.user),p(!0)):"SIGNED_OUT"===t&&(p(!1),i(null))});return()=>{t.subscription.unsubscribe()}},[h]),(0,r.jsx)(s.Provider,{value:{loggedIn:u,user:a,login:l,logout:d},children:e})}},28891:function(t,e,a){t.exports=a(98556)}},function(t){t.O(0,[540,640,215,620,304,339,954,470,730,744],function(){return t(t.s=38874)}),_N_E=t.O()}]);
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[958],{99166:function(e,t,s){Promise.resolve().then(s.bind(s,41400))},49532:function(e,t,s){"use strict";let n,i;async function r(){if(n||(i||(i=fetch("/api/config").then(e=>e.json()).then(e=>n=e.apiBaseUrl)),await i),void 0===n)throw Error("API base URL is undefined");return n}s.d(t,{b:function(){return r}})},41400:function(e,t,s){"use strict";s.r(t),s.d(t,{default:function(){return S}});var n=s(24004),i=s(14978),r=s(49532),l=s(49567),c=s(30469),a=s(65068),d=s(70417),o=s(21303),h=s(26485),u=s(30021),p=s(19708),x=s(65969),j=s(38640),v=s.n(j),m=s(28891),Z=s(40982),f=s(29794),w=s(22701),C=s(51956),y=s(82669),g=s(613),I=s(63147),k=s(34235);function P(e){var t,s,r,l;let{openDialog:p,handleClose:x,testCase:j}=e,[m,P]=(0,i.useState)(1);return(0,n.jsxs)(f.Z,{open:p,onClose:x,fullWidth:!0,maxWidth:"lg",children:[(0,n.jsxs)(y.Z,{children:["Dataset ",j.id.slice(0,6)]}),(0,n.jsxs)(C.Z,{children:[(0,n.jsx)(I.Z,{variant:"h6",style:{marginTop:"1rem"},children:"Test cases"}),(0,n.jsx)(g.u,{readOnly:!0,value:j&&k.default.dump(j.testCases),style:{width:"100%",padding:"0.75rem"},maxRows:15}),(0,n.jsx)(I.Z,{variant:"h6",style:{marginTop:"1rem"},children:"Used in..."}),(0,n.jsxs)(a.Z,{children:[(0,n.jsx)(h.Z,{children:(0,n.jsxs)(u.Z,{children:[(0,n.jsx)(o.Z,{children:"Eval ID"}),(0,n.jsx)(o.Z,{children:"Prompt ID"}),(0,n.jsx)(o.Z,{children:"Raw score"}),(0,n.jsx)(o.Z,{children:"Pass rate"}),(0,n.jsx)(o.Z,{children:"Pass count"}),(0,n.jsx)(o.Z,{children:"Fail count"}),(0,n.jsx)(o.Z,{children:"Prompt"})]})}),(0,n.jsx)(d.Z,{children:null==j?void 0:null===(t=j.prompts)||void 0===t?void 0:t.slice((m-1)*10,10*m).sort((e,t)=>t.evalId.localeCompare(e.evalId)).map((e,t)=>{var s,i,c,a,d;return(0,n.jsxs)(u.Z,{hover:!0,children:[(0,n.jsx)(o.Z,{children:(0,n.jsx)(v(),{href:"/eval/?evalId=".concat(e.evalId),children:e.evalId})}),(0,n.jsx)(o.Z,{style:{minWidth:"8em"},children:(0,n.jsx)(v(),{href:"/prompts/?id=".concat(e.id),children:e.id.slice(0,6)})}),(0,n.jsx)(o.Z,{children:"number"==typeof(null===(s=e.prompt.metrics)||void 0===s?void 0:s.score)?e.prompt.metrics.score.toFixed(2):"-"}),(0,n.jsx)(o.Z,{children:"number"==typeof(null===(i=e.prompt.metrics)||void 0===i?void 0:i.testPassCount)&&"number"==typeof(null===(c=e.prompt.metrics)||void 0===c?void 0:c.testFailCount)&&e.prompt.metrics.testPassCount+e.prompt.metrics.testFailCount>0?(e.prompt.metrics.testPassCount/(e.prompt.metrics.testPassCount+e.prompt.metrics.testFailCount)*100).toFixed(2)+"%":"-"}),(0,n.jsx)(o.Z,{children:null!==(r=null===(a=e.prompt.metrics)||void 0===a?void 0:a.testPassCount)&&void 0!==r?r:"-"}),(0,n.jsx)(o.Z,{children:null!==(l=null===(d=e.prompt.metrics)||void 0===d?void 0:d.testFailCount)&&void 0!==l?l:"-"}),(0,n.jsx)(o.Z,{children:e.prompt.raw.length>250?e.prompt.raw.slice(0,250)+"...":e.prompt.raw})]},t)})})]}),Math.ceil(((null==j?void 0:null===(s=j.prompts)||void 0===s?void 0:s.length)||0)/10)>1&&(0,n.jsx)(c.Z,{count:Math.ceil(j.prompts.length/10),page:m,onChange:(e,t)=>{P(t)}})]}),(0,n.jsx)(w.Z,{children:(0,n.jsx)(Z.Z,{onClick:x,children:"Close"})})]})}function S(){let e=(0,m.useSearchParams)(),[t,s]=(0,i.useState)([]),[j,Z]=(0,i.useState)("date"),[f,w]=(0,i.useState)("desc"),[C,y]=(0,i.useState)(1),[g,I]=(0,i.useState)(10),[k,S]=(0,i.useState)(!1),[b,E]=(0,i.useState)(0),D=e=>{let t=j===e&&"asc"===f?"desc":"asc";Z(e),w(t)};(0,i.useEffect)(()=>{(async()=>{fetch("".concat(await (0,r.b)(),"/api/datasets")).then(e=>e.json()).then(e=>{let t=[...e.data].sort((e,t)=>null===j?0:"asc"===f?e[j]>t[j]?1:-1:e[j]<t[j]?1:-1);s(t)})})()},[j,f,C,g]),(0,i.useEffect)(()=>{let s=null==e?void 0:e.get("id");if(s){let e=t.findIndex(e=>e.id.startsWith(s));-1!==e&&F(e)}},[t,e]);let F=e=>{E(e),S(!0)};return(0,n.jsxs)(l.Z,{paddingX:2,children:[(0,n.jsxs)(a.Z,{children:[(0,n.jsx)(h.Z,{children:(0,n.jsxs)(u.Z,{children:[(0,n.jsx)(o.Z,{style:{width:"10%"},children:"ID"}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:(0,n.jsx)(p.Z,{active:"raw"===j,direction:"raw"===j?f:"asc",onClick:()=>D("raw"),children:"Info"})}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:"Variables"}),(0,n.jsx)(o.Z,{style:{width:"10%"},children:(0,n.jsx)(p.Z,{active:"count"===j,direction:"count"===j?f:"asc",onClick:()=>D("count"),children:"Total # evals"})}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:(0,n.jsx)(x.Z,{title:"The date of the most recent eval for this set of test cases",children:(0,n.jsx)(p.Z,{active:"date"===j,direction:"date"===j?f:"asc",onClick:()=>D("date"),children:"Most recent eval date"})})}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:(0,n.jsx)(x.Z,{title:"The ID of the most recent eval for this set of test cases",children:(0,n.jsx)(p.Z,{active:"evalId"===j,direction:"evalId"===j?f:"asc",onClick:()=>D("evalId"),children:"Most recent eval ID"})})})]})}),(0,n.jsx)(d.Z,{children:t.slice((C-1)*g,C*g).map((e,t)=>(0,n.jsxs)(u.Z,{hover:!0,onClick:()=>F(t),style:{cursor:"pointer"},children:[(0,n.jsx)(o.Z,{children:e.id.slice(0,6)}),(0,n.jsxs)(o.Z,{style:{width:"20%",whiteSpace:"pre-wrap"},children:[e.testCases.length," test cases"]}),(0,n.jsx)(o.Z,{style:{width:"20%",whiteSpace:"pre-wrap"},children:(()=>{if(!Array.isArray(e.testCases)||"string"==typeof e.testCases[0])return"";let t=(e.testCases||[]).flatMap(e=>Object.keys(e.vars||{})),s=Array.from(new Set(t));return s.length>0?s.join(", "):"None"})()}),(0,n.jsx)(o.Z,{style:{width:"10%"},children:e.count}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:e.recentEvalDate||"Unknown"}),(0,n.jsx)(o.Z,{style:{width:"20%"},children:e.recentEvalId?(0,n.jsx)(v(),{href:"/eval?evalId=".concat(e.recentEvalId),children:e.recentEvalId}):"Unknown"})]},t))})]}),Math.ceil(t.length/g)>1&&(0,n.jsx)(c.Z,{count:Math.ceil(t.length/g),page:C,onChange:(e,t)=>y(t)}),t[b]&&(0,n.jsx)(P,{openDialog:k,handleClose:()=>{S(!1)},testCase:t[b]})]})}}},function(e){e.O(0,[540,640,215,770,995,374,235,322,470,730,744],function(){return e(e.s=99166)}),_N_E=e.O()}]);
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[185],{16741:function(e,t,n){Promise.resolve().then(n.bind(n,30224)),Promise.resolve().then(n.t.bind(n,82988,23)),Promise.resolve().then(n.t.bind(n,5100,23))},30224:function(e,t,n){"use strict";n.r(t),n.d(t,{PageShell:function(){return Z}});var r=n(24004),s=n(14978),o=n(52428),i=n(32414),a=n(38640),l=n.n(a),u=n(28891),c=n(95707),d=n(80378);function h(e){let{darkMode:t,onToggleDarkMode:n}=e;return(0,r.jsx)("div",{className:"dark-mode-toggle",onClick:n,children:t?(0,r.jsx)(c.Z,{}):(0,r.jsx)(d.Z,{})})}n(32854);var f=n(1445),m=n(31657),x=n(77656),g=n(36273),j=n(96976);function p(){let{user:e,logout:t}=(0,f.aC)(),[n,o]=s.useState(null),i=async()=>{null==t||t(),a()},a=()=>{o(null)};return e?(0,r.jsxs)("div",{children:[(0,r.jsx)(x.Z,{edge:"end","aria-label":"account of current user","aria-controls":"menu-appbar","aria-haspopup":"true",onClick:e=>{o(e.currentTarget)},color:"inherit",children:(0,r.jsx)(m.Z,{sx:{width:"1em",height:"1em",bgcolor:"#1976d2"}})}),(0,r.jsxs)(g.Z,{id:"menu-appbar",anchorEl:n,anchorOrigin:{vertical:"top",horizontal:"right"},keepMounted:!0,transformOrigin:{vertical:"top",horizontal:"right"},open:!!n,onClose:a,children:[(0,r.jsxs)(j.Z,{disabled:!0,children:["Logged in as ",e.email]}),(0,r.jsx)(j.Z,{onClick:i,children:"Logout"})]})]}):(0,r.jsx)(l(),{href:"/auth/signup/",children:(0,r.jsx)(x.Z,{edge:"end","aria-label":"User not logged in","aria-controls":"menu-appbar","aria-haspopup":"true",color:"inherit",children:(0,r.jsx)(m.Z,{sx:{width:"1em",height:"1em"}})})})}var v=n(49567),b=n(47051),E=n.n(b);function _(){return(0,r.jsxs)(v.Z,{className:"logo",children:[(0,r.jsx)(E(),{width:25,height:25,src:"/logo.svg",alt:"Promptfoo logo"})," ",(0,r.jsx)("span",{children:"promptfoo"})]})}n(17328),n(5008);var k=n(77580);function N(e){let{href:t,label:n}=e,s=(0,u.usePathname)()||"";return(0,r.jsx)(l(),{href:t,className:s.startsWith(t)?"active":"",children:n})}function O(e){let{darkMode:t,onToggleDarkMode:n}=e;return k.env.NEXT_PUBLIC_NO_BROWSING?(0,r.jsxs)(i.Z,{direction:"row",spacing:2,className:"nav",children:[(0,r.jsx)(_,{}),(0,r.jsx)(h,{darkMode:t,onToggleDarkMode:n})]}):(0,r.jsxs)(i.Z,{direction:"row",spacing:2,className:"nav",children:[(0,r.jsx)(_,{}),(0,r.jsx)(N,{href:"/setup",label:"New Eval"}),(0,r.jsx)(N,{href:"/eval",label:"Evals"}),(0,r.jsx)(N,{href:"/prompts",label:"Prompts"}),(0,r.jsx)(N,{href:"/datasets",label:"Datasets"}),(0,r.jsx)(N,{href:"/progress",label:"Progress"}),(0,r.jsxs)("div",{className:"right-aligned",children:[o.Ox?(0,r.jsx)(p,{}):null,(0,r.jsx)(h,{darkMode:t,onToggleDarkMode:n})]})]})}var C=n(51867),P=n(61191),S=n(38127);function Z(e){let{children:t}=e,n=(0,S.Z)("(prefers-color-scheme: dark)"),[o,i]=s.useState(n),a=s.useMemo(()=>(0,C.Z)({typography:{fontFamily:"inherit"},palette:{mode:o||n?"dark":"light"}}),[o,n]);return s.useEffect(()=>{n&&document.documentElement.setAttribute("data-theme","dark")},[n]),(0,r.jsx)(s.StrictMode,{children:(0,r.jsx)(P.Z,{theme:a,children:(0,r.jsx)(f.Ho,{children:(0,r.jsxs)(T,{children:[(0,r.jsx)(O,{darkMode:o,onToggleDarkMode:()=>{i(!o),o?document.documentElement.removeAttribute("data-theme"):document.documentElement.setAttribute("data-theme","dark")}}),(0,r.jsx)("div",{children:t})]})})})})}function T(e){let{children:t}=e;return(0,r.jsx)("div",{children:t})}n(70894)},52428:function(e,t,n){"use strict";n.d(t,{Ox:function(){return o},T8:function(){return s},eA:function(){return i}});var r=n(77580);let s=!r.env.NEXT_PUBLIC_PROMPTFOO_BUILD_STANDALONE_SERVER,o=!!r.env.NEXT_PUBLIC_PROMPTFOO_USE_SUPABASE,i=""},1445:function(e,t,n){"use strict";n.d(t,{Ho:function(){return d},OQ:function(){return i},aC:function(){return l}});var r=n(24004),s=n(14978),o=n(12594);let i=(0,o.createClientComponentClient)(),a=(0,s.createContext)({}),l=()=>(0,s.useContext)(a),u=(e,t)=>i.auth.signInWithPassword({email:e,password:t}),c=()=>i.auth.signOut(),d=e=>{let{children:t}=e,[n,o]=(0,s.useState)(null),[l,d]=(0,s.useState)(!1),h=(0,s.useCallback)(async()=>{let{data:e,error:t}=await i.auth.refreshSession();e&&o(e.user)},[]);return(0,s.useEffect)(()=>{h();let{data:e}=i.auth.onAuthStateChange(async(e,t)=>{"PASSWORD_RECOVERY"==e?d(!1):"SIGNED_IN"===e&&t?(o(t.user),d(!0)):"SIGNED_OUT"===e&&(d(!1),o(null))});return()=>{e.subscription.unsubscribe()}},[h]),(0,r.jsx)(a.Provider,{value:{loggedIn:l,user:n,login:u,logout:c},children:t})}},32854:function(){},17328:function(){},5008:function(){},70894:function(){},5100:function(){}},function(e){e.O(0,[540,640,620,954,166,470,730,744],function(){return e(e.s=16741)}),_N_E=e.O()}]);
@@ -1 +1 @@
1
- (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[92],{48931:function(e,t,a){"use strict";var i=a(71851);t.Z=void 0;var r=i(a(99570)),n=a(24004);t.Z=(0,r.default)((0,n.jsx)("path",{d:"M5 20h14v-2H5zM19 9h-4V3H9v6H5l7 7z"}),"Download")},96212:function(e,t,a){"use strict";a.d(t,{f:function(){return n}});var i=a(25800),r=a(46213);function n(e){return(0,r.ZP)("MuiListItemIcon",e)}let s=(0,i.Z)("MuiListItemIcon",["root","alignItemsFlexStart"]);t.Z=s},53975:function(e,t,a){"use strict";a.d(t,{L:function(){return n}});var i=a(25800),r=a(46213);function n(e){return(0,r.ZP)("MuiListItemText",e)}let s=(0,i.Z)("MuiListItemText",["root","multiline","dense","inset","primary","secondary"]);t.Z=s},96976:function(e,t,a){"use strict";a.d(t,{Z:function(){return M}});var i=a(94312),r=a(79042),n=a(14978),s=a(53468),o=a(3982),l=a(13477),d=a(31608),c=a(11874),u=a(89020),p=a(11112),v=a(30151),m=a(16656),h=a(41776),x=a(25800);let Z=(0,x.Z)("MuiDivider",["root","absolute","fullWidth","inset","middle","flexItem","light","vertical","withChildren","withChildrenVertical","textAlignRight","textAlignLeft","wrapper","wrapperVertical"]);var f=a(96212),g=a(53975),C=a(46213);function I(e){return(0,C.ZP)("MuiMenuItem",e)}let j=(0,x.Z)("MuiMenuItem",["root","focusVisible","dense","disabled","divider","gutters","selected"]);var b=a(24004);let y=["autoFocus","component","dense","divider","disableGutters","focusVisibleClassName","role","tabIndex","className"],P=e=>{let{disabled:t,dense:a,divider:i,disableGutters:n,selected:s,classes:l}=e,d=(0,o.Z)({root:["root",a&&"dense",t&&"disabled",!n&&"gutters",i&&"divider",s&&"selected"]},I,l);return(0,r.Z)({},l,d)},w=(0,d.ZP)(v.Z,{shouldForwardProp:e=>(0,c.Z)(e)||"classes"===e,name:"MuiMenuItem",slot:"Root",overridesResolver:(e,t)=>{let{ownerState:a}=e;return[t.root,a.dense&&t.dense,a.divider&&t.divider,!a.disableGutters&&t.gutters]}})(({theme:e,ownerState:t})=>(0,r.Z)({},e.typography.body1,{display:"flex",justifyContent:"flex-start",alignItems:"center",position:"relative",textDecoration:"none",minHeight:48,paddingTop:6,paddingBottom:6,boxSizing:"border-box",whiteSpace:"nowrap"},!t.disableGutters&&{paddingLeft:16,paddingRight:16},t.divider&&{borderBottom:`1px solid ${(e.vars||e).palette.divider}`,backgroundClip:"padding-box"},{"&:hover":{textDecoration:"none",backgroundColor:(e.vars||e).palette.action.hover,"@media (hover: none)":{backgroundColor:"transparent"}},[`&.${j.selected}`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / ${e.vars.palette.action.selectedOpacity})`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity),[`&.${j.focusVisible}`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / calc(${e.vars.palette.action.selectedOpacity} + ${e.vars.palette.action.focusOpacity}))`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity+e.palette.action.focusOpacity)}},[`&.${j.selected}:hover`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / calc(${e.vars.palette.action.selectedOpacity} + ${e.vars.palette.action.hoverOpacity}))`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity+e.palette.action.hoverOpacity),"@media (hover: none)":{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / ${e.vars.palette.action.selectedOpacity})`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity)}},[`&.${j.focusVisible}`]:{backgroundColor:(e.vars||e).palette.action.focus},[`&.${j.disabled}`]:{opacity:(e.vars||e).palette.action.disabledOpacity},[`& + .${Z.root}`]:{marginTop:e.spacing(1),marginBottom:e.spacing(1)},[`& + .${Z.inset}`]:{marginLeft:52},[`& .${g.Z.root}`]:{marginTop:0,marginBottom:0},[`& .${g.Z.inset}`]:{paddingLeft:36},[`& .${f.Z.root}`]:{minWidth:36}},!t.dense&&{[e.breakpoints.up("sm")]:{minHeight:"auto"}},t.dense&&(0,r.Z)({minHeight:32,paddingTop:4,paddingBottom:4},e.typography.body2,{[`& .${f.Z.root} svg`]:{fontSize:"1.25rem"}}))),k=n.forwardRef(function(e,t){let a;let o=(0,u.Z)({props:e,name:"MuiMenuItem"}),{autoFocus:l=!1,component:d="li",dense:c=!1,divider:v=!1,disableGutters:x=!1,focusVisibleClassName:Z,role:f="menuitem",tabIndex:g,className:C}=o,I=(0,i.Z)(o,y),j=n.useContext(p.Z),k=n.useMemo(()=>({dense:c||j.dense||!1,disableGutters:x}),[j.dense,c,x]),M=n.useRef(null);(0,m.Z)(()=>{l&&M.current&&M.current.focus()},[l]);let F=(0,r.Z)({},o,{dense:k.dense,divider:v,disableGutters:x}),S=P(o),$=(0,h.Z)(M,t);return o.disabled||(a=void 0!==g?g:-1),(0,b.jsx)(p.Z.Provider,{value:k,children:(0,b.jsx)(w,(0,r.Z)({ref:$,role:f,tabIndex:a,component:d,focusVisibleClassName:(0,s.Z)(S.focusVisible,Z),className:(0,s.Z)(S.root,C)},I,{ownerState:F,classes:S}))})});var M=k},11340:function(e,t,a){Promise.resolve().then(a.bind(a,19259))},19259:function(e,t,a){"use strict";a.r(t),a.d(t,{default:function(){return I}});var i=a(24004),r=a(14978),n=a(38640),s=a.n(n),o=a(68873),l=a(49567),d=a(40982),c=a(48931),u=a(36273),p=a(96976),v=a(30469),m=a(65068),h=a(70417),x=a(21303),Z=a(26485),f=a(30021),g=a(19708),C=a(67339);function I(){let[e,t]=(0,r.useState)([]),[a,n]=(0,r.useState)(null),[I,j]=(0,r.useState)("asc"),[b,y]=(0,r.useState)(null),[P,w]=r.useState(1),[k,M]=(0,r.useState)({evalId:"",datasetId:"",provider:"",promptId:""}),F=!!b;(0,r.useEffect)(()=>{(async()=>{let e=await fetch("/api/progress"),a=await e.json();a&&a.data&&t(a.data)})()},[]);let S=e=>{let t=a===e&&"asc"===I;n(e),j(t?"desc":"asc")},$=t=>{let a="json"===t?JSON.stringify(e):R(e),i=new Blob([a],{type:"text/".concat(t,";charset=utf-8;")}),r=document.createElement("a");r.href=URL.createObjectURL(i),r.download="cols_export.".concat(t),r.click(),URL.revokeObjectURL(r.href),y(null)},O=e=>(null==e?void 0:e.testPassCount)!=null&&(null==e?void 0:e.testFailCount)!=null?(e.testPassCount/(e.testPassCount+e.testFailCount)*100).toFixed(2):"-",R=e=>{var t,a,i,r,n;let s=e.map(e=>{var s,o,l,d,c;return[null!==(t=e.evalId)&&void 0!==t?t:"",null!==(a=null===(s=e.datasetId)||void 0===s?void 0:s.slice(0,6))&&void 0!==a?a:"",null!==(i=e.provider)&&void 0!==i?i:"",(null!==(r=null===(o=e.promptId)||void 0===o?void 0:o.slice(0,6))&&void 0!==r?r:"")+" "+(null!==(n=e.raw)&&void 0!==n?n:""),O(e.metrics),(null===(l=e.metrics)||void 0===l?void 0:l.testPassCount)==null?"-":"".concat(e.metrics.testPassCount),(null===(d=e.metrics)||void 0===d?void 0:d.testFailCount)==null?"-":"".concat(e.metrics.testFailCount),(null===(c=e.metrics)||void 0===c?void 0:c.score)==null?"-":e.metrics.score.toFixed(2)]});return[["Eval","Dataset","Provider","Prompt","Pass Rate %","Pass Count","Fail Count","Raw score"]].concat(s).map(e=>e.map(e=>null!=e?e:"").join(",")).join("\n")},L=r.useMemo(()=>e.filter(e=>{var t,a,i,r;return(!k.evalId||(null===(t=e.evalId)||void 0===t?void 0:t.includes(k.evalId)))&&(!k.datasetId||(null===(a=e.datasetId)||void 0===a?void 0:a.startsWith(k.datasetId)))&&(!k.provider||(null===(i=e.provider)||void 0===i?void 0:i.includes(k.provider)))&&(!k.promptId||(null===(r=e.promptId)||void 0===r?void 0:r.startsWith(k.promptId)))}),[e,k]),D=r.useMemo(()=>L.sort((e,t)=>{if(!a)return 0;if("passRate"===a){let a=parseFloat(O(e.metrics)),i=parseFloat(O(t.metrics));return"asc"===I?a-i:i-a}if(a in e&&a in t){let i=e[a]||"",r=t[a]||"";return"asc"===I?i.toString().localeCompare(r.toString()):r.toString().localeCompare(i.toString())}return 0}),[L,a,I]),E=r.useMemo(()=>Array.from(new Set(e.map(e=>e.evalId))),[e]),V=r.useMemo(()=>Array.from(new Set(e.map(e=>e.datasetId))),[e]),N=r.useMemo(()=>Array.from(new Set(e.map(e=>e.provider))),[e]),z=r.useMemo(()=>Array.from(new Set(e.map(e=>e.promptId))),[e]);return(0,i.jsxs)(l.Z,{paddingX:2,children:[(0,i.jsxs)(l.Z,{display:"flex",justifyContent:"space-between",alignItems:"center",children:[(0,i.jsx)("h2",{children:"Progress summary"}),(0,i.jsxs)("div",{children:[(0,i.jsx)(d.Z,{id:"export-button","aria-controls":F?"export-menu":void 0,"aria-haspopup":"true","aria-expanded":F?"true":void 0,onClick:e=>{y(e.currentTarget)},startIcon:(0,i.jsx)(c.Z,{}),children:"Export"}),(0,i.jsxs)(u.Z,{id:"export-menu",anchorEl:b,open:F,onClose:()=>{y(null)},MenuListProps:{"aria-labelledby":"export-button"},children:[(0,i.jsx)(p.Z,{onClick:()=>$("csv"),children:"CSV"}),(0,i.jsx)(p.Z,{onClick:()=>$("json"),children:"JSON"})]})]})]}),(0,i.jsx)(l.Z,{children:"This page shows performance metrics for recent evals."}),(0,i.jsxs)(l.Z,{display:"flex",flexDirection:"row",gap:2,mt:2,children:[(0,i.jsx)(o.Z,{options:E,value:k.evalId,onChange:(e,t)=>{M({...k,evalId:t||""})},renderInput:e=>(0,i.jsx)(C.Z,{...e,label:"Eval ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(o.Z,{options:V,value:k.datasetId,onChange:(e,t)=>{M({...k,datasetId:t||""})},renderInput:e=>(0,i.jsx)(C.Z,{...e,label:"Dataset ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(o.Z,{options:N,value:k.provider,onChange:(e,t)=>{M({...k,provider:t||""})},renderInput:e=>(0,i.jsx)(C.Z,{...e,label:"Provider",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(o.Z,{options:z,value:k.promptId,onChange:(e,t)=>{M({...k,promptId:t||""})},renderInput:e=>(0,i.jsx)(C.Z,{...e,label:"Prompt ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}})]}),(0,i.jsxs)(m.Z,{children:[(0,i.jsx)(Z.Z,{children:(0,i.jsxs)(f.Z,{children:[(0,i.jsx)(x.Z,{children:(0,i.jsx)(g.Z,{active:"evalId"===a,direction:"evalId"===a?I:"asc",onClick:()=>S("evalId"),children:"Eval"})}),(0,i.jsx)(x.Z,{children:"Dataset"}),(0,i.jsx)(x.Z,{children:"Provider"}),(0,i.jsx)(x.Z,{children:"Prompt"}),(0,i.jsx)(x.Z,{children:(0,i.jsx)(g.Z,{active:"passRate"===a,direction:"passRate"===a?I:"asc",onClick:()=>S("passRate"),children:"Pass Rate %"})}),(0,i.jsx)(x.Z,{children:(0,i.jsx)(g.Z,{active:"testPassCount"===a,direction:"testPassCount"===a?I:"asc",onClick:()=>S("testPassCount"),children:"Pass Count"})}),(0,i.jsx)(x.Z,{children:(0,i.jsx)(g.Z,{active:"testFailCount"===a,direction:"testFailCount"===a?I:"asc",onClick:()=>S("testFailCount"),children:"Fail Count"})}),(0,i.jsx)(x.Z,{children:(0,i.jsx)(g.Z,{active:"score"===a,direction:"score"===a?I:"asc",onClick:()=>S("score"),children:"Raw score"})})]})}),(0,i.jsx)(h.Z,{children:D.slice((P-1)*25,25*P).map((e,t)=>{var a,r,n,o,l;return(0,i.jsxs)(f.Z,{hover:!0,onClick:()=>M({...k,evalId:e.evalId,datasetId:e.datasetId||"",promptId:e.promptId||"",provider:e.provider}),children:[(0,i.jsx)(x.Z,{children:(0,i.jsx)(s(),{href:"/eval?evalId=".concat(e.evalId),onClick:e=>e.stopPropagation(),children:e.evalId})}),(0,i.jsx)(x.Z,{children:(0,i.jsx)(s(),{href:"/datasets?id=".concat(e.datasetId),onClick:e=>e.stopPropagation(),children:null===(a=e.datasetId)||void 0===a?void 0:a.slice(0,6)})}),(0,i.jsx)(x.Z,{children:e.provider}),(0,i.jsxs)(x.Z,{children:[(0,i.jsxs)(s(),{href:"/prompts?id=".concat(e.promptId),onClick:e=>e.stopPropagation(),children:["[",null===(r=e.promptId)||void 0===r?void 0:r.slice(0,6),"]"]})," ",e.raw]}),(0,i.jsx)(x.Z,{children:O(e.metrics)}),(0,i.jsx)(x.Z,{children:(null===(n=e.metrics)||void 0===n?void 0:n.testPassCount)==null?"-":"".concat(e.metrics.testPassCount)}),(0,i.jsx)(x.Z,{children:(null===(o=e.metrics)||void 0===o?void 0:o.testFailCount)==null?"-":"".concat(e.metrics.testFailCount)}),(0,i.jsx)(x.Z,{children:(null===(l=e.metrics)||void 0===l?void 0:l.score)==null?"-":e.metrics.score.toFixed(2)})]},t)})})]}),Math.ceil(L.length/25)>1&&(0,i.jsx)(v.Z,{count:Math.ceil(D.length/25),page:P,onChange:(e,t)=>w(t),sx:{pt:2,pb:4,display:"flex",justifyContent:"center"}})]})}}},function(e){e.O(0,[540,640,215,620,770,304,339,378,873,322,470,730,744],function(){return e(e.s=11340)}),_N_E=e.O()}]);
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[92],{48931:function(e,t,a){"use strict";var i=a(71851);t.Z=void 0;var r=i(a(99570)),n=a(24004);t.Z=(0,r.default)((0,n.jsx)("path",{d:"M5 20h14v-2H5zM19 9h-4V3H9v6H5l7 7z"}),"Download")},96212:function(e,t,a){"use strict";a.d(t,{f:function(){return n}});var i=a(25800),r=a(46213);function n(e){return(0,r.ZP)("MuiListItemIcon",e)}let s=(0,i.Z)("MuiListItemIcon",["root","alignItemsFlexStart"]);t.Z=s},53975:function(e,t,a){"use strict";a.d(t,{L:function(){return n}});var i=a(25800),r=a(46213);function n(e){return(0,r.ZP)("MuiListItemText",e)}let s=(0,i.Z)("MuiListItemText",["root","multiline","dense","inset","primary","secondary"]);t.Z=s},96976:function(e,t,a){"use strict";a.d(t,{Z:function(){return M}});var i=a(94312),r=a(79042),n=a(14978),s=a(53468),o=a(3982),l=a(13477),d=a(31608),c=a(11874),u=a(89020),p=a(11112),v=a(30151),m=a(16656),h=a(41776),x=a(25800);let Z=(0,x.Z)("MuiDivider",["root","absolute","fullWidth","inset","middle","flexItem","light","vertical","withChildren","withChildrenVertical","textAlignRight","textAlignLeft","wrapper","wrapperVertical"]);var f=a(96212),g=a(53975),C=a(46213);function I(e){return(0,C.ZP)("MuiMenuItem",e)}let j=(0,x.Z)("MuiMenuItem",["root","focusVisible","dense","disabled","divider","gutters","selected"]);var b=a(24004);let y=["autoFocus","component","dense","divider","disableGutters","focusVisibleClassName","role","tabIndex","className"],P=e=>{let{disabled:t,dense:a,divider:i,disableGutters:n,selected:s,classes:l}=e,d=(0,o.Z)({root:["root",a&&"dense",t&&"disabled",!n&&"gutters",i&&"divider",s&&"selected"]},I,l);return(0,r.Z)({},l,d)},w=(0,d.ZP)(v.Z,{shouldForwardProp:e=>(0,c.Z)(e)||"classes"===e,name:"MuiMenuItem",slot:"Root",overridesResolver:(e,t)=>{let{ownerState:a}=e;return[t.root,a.dense&&t.dense,a.divider&&t.divider,!a.disableGutters&&t.gutters]}})(({theme:e,ownerState:t})=>(0,r.Z)({},e.typography.body1,{display:"flex",justifyContent:"flex-start",alignItems:"center",position:"relative",textDecoration:"none",minHeight:48,paddingTop:6,paddingBottom:6,boxSizing:"border-box",whiteSpace:"nowrap"},!t.disableGutters&&{paddingLeft:16,paddingRight:16},t.divider&&{borderBottom:`1px solid ${(e.vars||e).palette.divider}`,backgroundClip:"padding-box"},{"&:hover":{textDecoration:"none",backgroundColor:(e.vars||e).palette.action.hover,"@media (hover: none)":{backgroundColor:"transparent"}},[`&.${j.selected}`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / ${e.vars.palette.action.selectedOpacity})`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity),[`&.${j.focusVisible}`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / calc(${e.vars.palette.action.selectedOpacity} + ${e.vars.palette.action.focusOpacity}))`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity+e.palette.action.focusOpacity)}},[`&.${j.selected}:hover`]:{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / calc(${e.vars.palette.action.selectedOpacity} + ${e.vars.palette.action.hoverOpacity}))`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity+e.palette.action.hoverOpacity),"@media (hover: none)":{backgroundColor:e.vars?`rgba(${e.vars.palette.primary.mainChannel} / ${e.vars.palette.action.selectedOpacity})`:(0,l.Fq)(e.palette.primary.main,e.palette.action.selectedOpacity)}},[`&.${j.focusVisible}`]:{backgroundColor:(e.vars||e).palette.action.focus},[`&.${j.disabled}`]:{opacity:(e.vars||e).palette.action.disabledOpacity},[`& + .${Z.root}`]:{marginTop:e.spacing(1),marginBottom:e.spacing(1)},[`& + .${Z.inset}`]:{marginLeft:52},[`& .${g.Z.root}`]:{marginTop:0,marginBottom:0},[`& .${g.Z.inset}`]:{paddingLeft:36},[`& .${f.Z.root}`]:{minWidth:36}},!t.dense&&{[e.breakpoints.up("sm")]:{minHeight:"auto"}},t.dense&&(0,r.Z)({minHeight:32,paddingTop:4,paddingBottom:4},e.typography.body2,{[`& .${f.Z.root} svg`]:{fontSize:"1.25rem"}}))),k=n.forwardRef(function(e,t){let a;let o=(0,u.Z)({props:e,name:"MuiMenuItem"}),{autoFocus:l=!1,component:d="li",dense:c=!1,divider:v=!1,disableGutters:x=!1,focusVisibleClassName:Z,role:f="menuitem",tabIndex:g,className:C}=o,I=(0,i.Z)(o,y),j=n.useContext(p.Z),k=n.useMemo(()=>({dense:c||j.dense||!1,disableGutters:x}),[j.dense,c,x]),M=n.useRef(null);(0,m.Z)(()=>{l&&M.current&&M.current.focus()},[l]);let F=(0,r.Z)({},o,{dense:k.dense,divider:v,disableGutters:x}),S=P(o),$=(0,h.Z)(M,t);return o.disabled||(a=void 0!==g?g:-1),(0,b.jsx)(p.Z.Provider,{value:k,children:(0,b.jsx)(w,(0,r.Z)({ref:$,role:f,tabIndex:a,component:d,focusVisibleClassName:(0,s.Z)(S.focusVisible,Z),className:(0,s.Z)(S.root,C)},I,{ownerState:F,classes:S}))})});var M=k},11340:function(e,t,a){Promise.resolve().then(a.bind(a,19259))},19259:function(e,t,a){"use strict";a.r(t),a.d(t,{default:function(){return I}});var i=a(24004),r=a(14978),n=a(48931),s=a(68873),o=a(49567),l=a(40982),d=a(36273),c=a(96976),u=a(30469),p=a(65068),v=a(70417),m=a(21303),h=a(26485),x=a(30021),Z=a(19708),f=a(67339),g=a(38640),C=a.n(g);function I(){let[e,t]=(0,r.useState)([]),[a,g]=(0,r.useState)(null),[I,j]=(0,r.useState)("asc"),[b,y]=(0,r.useState)(null),[P,w]=r.useState(1),[k,M]=(0,r.useState)({evalId:"",datasetId:"",provider:"",promptId:""}),F=!!b;(0,r.useEffect)(()=>{(async()=>{let e=await fetch("/api/progress"),a=await e.json();a&&a.data&&t(a.data)})()},[]);let S=e=>{let t=a===e&&"asc"===I;g(e),j(t?"desc":"asc")},$=t=>{let a="json"===t?JSON.stringify(e):R(e),i=new Blob([a],{type:"text/".concat(t,";charset=utf-8;")}),r=document.createElement("a");r.href=URL.createObjectURL(i),r.download="cols_export.".concat(t),r.click(),URL.revokeObjectURL(r.href),y(null)},O=e=>(null==e?void 0:e.testPassCount)!=null&&(null==e?void 0:e.testFailCount)!=null?(e.testPassCount/(e.testPassCount+e.testFailCount)*100).toFixed(2):"-",R=e=>{var t,a,i,r,n;let s=e.map(e=>{var s,o,l,d,c;return[null!==(t=e.evalId)&&void 0!==t?t:"",null!==(a=null===(s=e.datasetId)||void 0===s?void 0:s.slice(0,6))&&void 0!==a?a:"",null!==(i=e.provider)&&void 0!==i?i:"",(null!==(r=null===(o=e.promptId)||void 0===o?void 0:o.slice(0,6))&&void 0!==r?r:"")+" "+(null!==(n=e.raw)&&void 0!==n?n:""),O(e.metrics),(null===(l=e.metrics)||void 0===l?void 0:l.testPassCount)==null?"-":"".concat(e.metrics.testPassCount),(null===(d=e.metrics)||void 0===d?void 0:d.testFailCount)==null?"-":"".concat(e.metrics.testFailCount),(null===(c=e.metrics)||void 0===c?void 0:c.score)==null?"-":e.metrics.score.toFixed(2)]});return[["Eval","Dataset","Provider","Prompt","Pass Rate %","Pass Count","Fail Count","Raw score"]].concat(s).map(e=>e.map(e=>null!=e?e:"").join(",")).join("\n")},L=r.useMemo(()=>e.filter(e=>{var t,a,i,r;return(!k.evalId||(null===(t=e.evalId)||void 0===t?void 0:t.includes(k.evalId)))&&(!k.datasetId||(null===(a=e.datasetId)||void 0===a?void 0:a.startsWith(k.datasetId)))&&(!k.provider||(null===(i=e.provider)||void 0===i?void 0:i.includes(k.provider)))&&(!k.promptId||(null===(r=e.promptId)||void 0===r?void 0:r.startsWith(k.promptId)))}),[e,k]),D=r.useMemo(()=>L.sort((e,t)=>{if(!a)return 0;if("passRate"===a){let a=parseFloat(O(e.metrics)),i=parseFloat(O(t.metrics));return"asc"===I?a-i:i-a}if(a in e&&a in t){let i=e[a]||"",r=t[a]||"";return"asc"===I?i.toString().localeCompare(r.toString()):r.toString().localeCompare(i.toString())}return 0}),[L,a,I]),E=r.useMemo(()=>Array.from(new Set(e.map(e=>e.evalId))),[e]),V=r.useMemo(()=>Array.from(new Set(e.map(e=>e.datasetId))),[e]),N=r.useMemo(()=>Array.from(new Set(e.map(e=>e.provider))),[e]),z=r.useMemo(()=>Array.from(new Set(e.map(e=>e.promptId))),[e]);return(0,i.jsxs)(o.Z,{paddingX:2,children:[(0,i.jsxs)(o.Z,{display:"flex",justifyContent:"space-between",alignItems:"center",children:[(0,i.jsx)("h2",{children:"Progress summary"}),(0,i.jsxs)("div",{children:[(0,i.jsx)(l.Z,{id:"export-button","aria-controls":F?"export-menu":void 0,"aria-haspopup":"true","aria-expanded":F?"true":void 0,onClick:e=>{y(e.currentTarget)},startIcon:(0,i.jsx)(n.Z,{}),children:"Export"}),(0,i.jsxs)(d.Z,{id:"export-menu",anchorEl:b,open:F,onClose:()=>{y(null)},MenuListProps:{"aria-labelledby":"export-button"},children:[(0,i.jsx)(c.Z,{onClick:()=>$("csv"),children:"CSV"}),(0,i.jsx)(c.Z,{onClick:()=>$("json"),children:"JSON"})]})]})]}),(0,i.jsx)(o.Z,{children:"This page shows performance metrics for recent evals."}),(0,i.jsxs)(o.Z,{display:"flex",flexDirection:"row",gap:2,mt:2,children:[(0,i.jsx)(s.Z,{options:E,value:k.evalId,onChange:(e,t)=>{M({...k,evalId:t||""})},renderInput:e=>(0,i.jsx)(f.Z,{...e,label:"Eval ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(s.Z,{options:V,value:k.datasetId,onChange:(e,t)=>{M({...k,datasetId:t||""})},renderInput:e=>(0,i.jsx)(f.Z,{...e,label:"Dataset ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(s.Z,{options:N,value:k.provider,onChange:(e,t)=>{M({...k,provider:t||""})},renderInput:e=>(0,i.jsx)(f.Z,{...e,label:"Provider",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}}),(0,i.jsx)(s.Z,{options:z,value:k.promptId,onChange:(e,t)=>{M({...k,promptId:t||""})},renderInput:e=>(0,i.jsx)(f.Z,{...e,label:"Prompt ID",variant:"outlined",size:"small",fullWidth:!0}),sx:{width:220}})]}),(0,i.jsxs)(p.Z,{children:[(0,i.jsx)(h.Z,{children:(0,i.jsxs)(x.Z,{children:[(0,i.jsx)(m.Z,{children:(0,i.jsx)(Z.Z,{active:"evalId"===a,direction:"evalId"===a?I:"asc",onClick:()=>S("evalId"),children:"Eval"})}),(0,i.jsx)(m.Z,{children:"Dataset"}),(0,i.jsx)(m.Z,{children:"Provider"}),(0,i.jsx)(m.Z,{children:"Prompt"}),(0,i.jsx)(m.Z,{children:(0,i.jsx)(Z.Z,{active:"passRate"===a,direction:"passRate"===a?I:"asc",onClick:()=>S("passRate"),children:"Pass Rate %"})}),(0,i.jsx)(m.Z,{children:(0,i.jsx)(Z.Z,{active:"testPassCount"===a,direction:"testPassCount"===a?I:"asc",onClick:()=>S("testPassCount"),children:"Pass Count"})}),(0,i.jsx)(m.Z,{children:(0,i.jsx)(Z.Z,{active:"testFailCount"===a,direction:"testFailCount"===a?I:"asc",onClick:()=>S("testFailCount"),children:"Fail Count"})}),(0,i.jsx)(m.Z,{children:(0,i.jsx)(Z.Z,{active:"score"===a,direction:"score"===a?I:"asc",onClick:()=>S("score"),children:"Raw score"})})]})}),(0,i.jsx)(v.Z,{children:D.slice((P-1)*25,25*P).map((e,t)=>{var a,r,n,s,o;return(0,i.jsxs)(x.Z,{hover:!0,onClick:()=>M({...k,evalId:e.evalId,datasetId:e.datasetId||"",promptId:e.promptId||"",provider:e.provider}),children:[(0,i.jsx)(m.Z,{children:(0,i.jsx)(C(),{href:"/eval?evalId=".concat(e.evalId),onClick:e=>e.stopPropagation(),children:e.evalId})}),(0,i.jsx)(m.Z,{children:(0,i.jsx)(C(),{href:"/datasets?id=".concat(e.datasetId),onClick:e=>e.stopPropagation(),children:null===(a=e.datasetId)||void 0===a?void 0:a.slice(0,6)})}),(0,i.jsx)(m.Z,{children:e.provider}),(0,i.jsxs)(m.Z,{children:[(0,i.jsxs)(C(),{href:"/prompts?id=".concat(e.promptId),onClick:e=>e.stopPropagation(),children:["[",null===(r=e.promptId)||void 0===r?void 0:r.slice(0,6),"]"]})," ",e.raw]}),(0,i.jsx)(m.Z,{children:O(e.metrics)}),(0,i.jsx)(m.Z,{children:(null===(n=e.metrics)||void 0===n?void 0:n.testPassCount)==null?"-":"".concat(e.metrics.testPassCount)}),(0,i.jsx)(m.Z,{children:(null===(s=e.metrics)||void 0===s?void 0:s.testFailCount)==null?"-":"".concat(e.metrics.testFailCount)}),(0,i.jsx)(m.Z,{children:(null===(o=e.metrics)||void 0===o?void 0:o.score)==null?"-":e.metrics.score.toFixed(2)})]},t)})})]}),Math.ceil(L.length/25)>1&&(0,i.jsx)(u.Z,{count:Math.ceil(D.length/25),page:P,onChange:(e,t)=>w(t),sx:{pt:2,pb:4,display:"flex",justifyContent:"center"}})]})}}},function(e){e.O(0,[540,640,215,620,770,304,339,378,873,322,470,730,744],function(){return e(e.s=11340)}),_N_E=e.O()}]);
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[794],{38486:function(e,t,n){Promise.resolve().then(n.bind(n,73730))},49532:function(e,t,n){"use strict";let i,s;async function l(){if(i||(s||(s=fetch("/api/config").then(e=>e.json()).then(e=>i=e.apiBaseUrl)),await s),void 0===i)throw Error("API base URL is undefined");return i}n.d(t,{b:function(){return l}})},73730:function(e,t,n){"use strict";n.r(t),n.d(t,{default:function(){return I}});var i=n(24004),s=n(14978),l=n(49532),r=n(49567),c=n(30469),d=n(65068),a=n(70417),o=n(21303),h=n(26485),u=n(30021),x=n(19708),j=n(65969),Z=n(38640),v=n.n(Z),p=n(28891),f=n(40982),m=n(29794),w=n(22701),y=n(51956),C=n(82669),g=n(613),E=n(63147),k=e=>{var t;let{openDialog:n,handleClose:s,selectedPrompt:l}=e;return(0,i.jsxs)(m.Z,{open:n,onClose:s,fullWidth:!0,maxWidth:"lg",children:[(0,i.jsxs)(C.Z,{children:["Prompt ",l.id.slice(0,6)]}),(0,i.jsxs)(y.Z,{children:[(0,i.jsx)(E.Z,{variant:"h6",style:{marginTop:"1rem"},children:"Prompt"}),(0,i.jsx)(g.u,{readOnly:!0,value:null==l?void 0:null===(t=l.prompt)||void 0===t?void 0:t.raw,style:{width:"100%",padding:"0.75rem"},maxRows:50}),(0,i.jsx)(E.Z,{variant:"h6",style:{marginTop:"1rem"},children:"Used in..."}),(0,i.jsxs)(d.Z,{children:[(0,i.jsx)(h.Z,{children:(0,i.jsxs)(u.Z,{children:[(0,i.jsx)(o.Z,{children:"Eval ID"}),(0,i.jsx)(o.Z,{children:"Dataset ID"}),(0,i.jsx)(o.Z,{children:"Raw score"}),(0,i.jsx)(o.Z,{children:"Pass rate"}),(0,i.jsx)(o.Z,{children:"Pass count"}),(0,i.jsx)(o.Z,{children:"Fail count"})]})}),(0,i.jsx)(a.Z,{children:null==l?void 0:l.evals.sort((e,t)=>t.id.localeCompare(e.id)).map(e=>{var t,n,s,l,r,c;let d=null!==(l=null===(t=e.metrics)||void 0===t?void 0:t.testPassCount)&&void 0!==l?l:0,a=null!==(r=null===(n=e.metrics)||void 0===n?void 0:n.testFailCount)&&void 0!==r?r:0,h=d+a>0?(d/(d+a)*100).toFixed(2)+"%":"-";return(0,i.jsxs)(u.Z,{children:[(0,i.jsx)(o.Z,{children:(0,i.jsx)(v(),{href:"/eval/?evalId=".concat(e.id),children:e.id})}),(0,i.jsx)(o.Z,{children:(0,i.jsx)(v(),{href:"/datasets/?id=".concat(e.datasetId),children:e.datasetId.slice(0,6)})}),(0,i.jsx)(o.Z,{children:null!==(c=null===(s=e.metrics)||void 0===s?void 0:s.score.toFixed(2))&&void 0!==c?c:"-"}),(0,i.jsx)(o.Z,{children:h}),(0,i.jsx)(o.Z,{children:d}),(0,i.jsx)(o.Z,{children:a})]},"eval-".concat(e.id))})})]})]}),(0,i.jsx)(w.Z,{children:(0,i.jsx)(f.Z,{onClick:s,children:"Close"})})]})};function I(){let e=(0,p.useSearchParams)(),[t,n]=(0,s.useState)([]),[Z,f]=(0,s.useState)("date"),[m,w]=(0,s.useState)("desc"),[y,C]=(0,s.useState)(1),[g,E]=(0,s.useState)(10),[I,P]=(0,s.useState)(!1),[S,D]=(0,s.useState)(0),b=e=>{let t=Z===e&&"asc"===m?"desc":"asc";f(e),w(t)};(0,s.useEffect)(()=>{(async()=>{fetch("".concat(await (0,l.b)(),"/api/prompts")).then(e=>e.json()).then(e=>{let t=[...e.data].sort((e,t)=>null===Z?0:"asc"===m?e[Z]>t[Z]?1:-1:e[Z]<t[Z]?1:-1);n(t)})})()},[Z,m]),(0,s.useEffect)(()=>{let n=null==e?void 0:e.get("id");if(n){let e=t.findIndex(e=>e.id.startsWith(n));-1!==e&&_(e)}},[t,e]);let _=e=>{P(!0),D(e)};return(0,i.jsxs)(r.Z,{paddingX:2,children:[(0,i.jsxs)(d.Z,{children:[(0,i.jsx)(h.Z,{children:(0,i.jsxs)(u.Z,{children:[(0,i.jsx)(o.Z,{style:{width:"10%"},children:"ID"}),(0,i.jsx)(o.Z,{style:{width:"60%"},children:(0,i.jsx)(x.Z,{active:"raw"===Z,direction:"raw"===Z?m:"asc",onClick:()=>b("raw"),children:"Prompt"})}),(0,i.jsx)(o.Z,{style:{width:"20%"},children:(0,i.jsx)(j.Z,{title:"The date of the most recent eval for this prompt",children:(0,i.jsx)(x.Z,{active:"date"===Z,direction:"date"===Z?m:"asc",onClick:()=>b("date"),children:"Most recent eval"})})}),(0,i.jsx)(o.Z,{style:{width:"10%"},children:(0,i.jsx)(x.Z,{active:"count"===Z,direction:"count"===Z?m:"asc",onClick:()=>b("count"),children:"# Evals"})})]})}),(0,i.jsx)(a.Z,{children:t.slice((y-1)*g,y*g).map((e,t)=>(0,i.jsxs)(u.Z,{hover:!0,children:[(0,i.jsx)(o.Z,{style:{width:"10%"},children:e.id.slice(0,6)}),(0,i.jsx)(o.Z,{style:{width:"60%",whiteSpace:"pre-wrap",cursor:"pointer"},onClick:()=>_(t),children:e.prompt.raw.length>500?e.prompt.raw.slice(0,500)+"...":e.prompt.raw}),(0,i.jsx)(o.Z,{style:{width:"20%"},children:e.recentEvalDate?(0,i.jsx)(v(),{href:"/eval?evalId=".concat(e.recentEvalId),children:e.recentEvalDate}):"Unknown"}),(0,i.jsx)(o.Z,{style:{width:"10%"},children:e.count})]},t))})]}),Math.ceil(t.length/g)>1&&(0,i.jsx)(c.Z,{count:Math.ceil(t.length/g),page:y,onChange:(e,t)=>C(t)}),t[S]&&(0,i.jsx)(k,{openDialog:I,handleClose:()=>{P(!1)},selectedPrompt:t[S]})]})}}},function(e){e.O(0,[540,640,215,770,995,374,322,470,730,744],function(){return e(e.s=38486)}),_N_E=e.O()}]);
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[420],{25902:function(e,i,t){Promise.resolve().then(t.bind(t,37493))},49532:function(e,i,t){"use strict";let a,r;async function s(){if(a||(r||(r=fetch("/api/config").then(e=>e.json()).then(e=>a=e.apiBaseUrl)),await r),void 0===a)throw Error("API base URL is undefined");return a}t.d(i,{b:function(){return s}})},37493:function(e,i,t){"use strict";t.r(i),t.d(i,{default:function(){return G}});var a,r,s=t(24004),n=t(14978),l=t(49532),o=t(885),c=t(11520),m=t(93650),u=t(63147),h=t(80661),d=t(49578);let p={"Brand Risk":["competitors","politics","excessive-agency","hallucination","overreliance","harmful:graphic-content","harmful:indiscriminate-weapons","harmful:misinformation-disinformation","harmful:non-violent-crime","harmful:radicalization","harmful:unsafe-practices"],"Legal Risk":["harmful:child-exploitation","harmful:copyright-violations","harmful:cybercrime","harmful:illegal-activities","harmful:illegal-drugs","harmful:intellectual-property","harmful:privacy","harmful:sex-crime","harmful:sexual-content","harmful:specialized-advice","harmful:violent-crime","harmful:self-harm","contracts"],"Malicious Attacks":["harmful:harassment-bullying","harmful:hate","harmful:insults","harmful:profanity","hijacking","jailbreak","pii","prompt-injection"]},f={"Brand Risk":"Risks that can affect the brand reputation and trustworthiness.","Legal Risk":"Risks that can lead to legal consequences or violations.","Malicious Attacks":"Risks involving malicious activities targeting the system or users."},g={"excessive-agency":"Model taking excessive initiative or misunderstanding its capabilities.","harmful:child-exploitation":"Content exploiting or harming children.","harmful:copyright-violations":"Content violating copyright laws.","harmful:cybercrime":"Content related to cybercriminal activities.","harmful:graphic-content":"Displaying graphic or violent content.","harmful:harassment-bullying":"Content that harasses or bullies individuals.","harmful:hate":"Content that promotes hate or discrimination.","harmful:illegal-activities":"Content promoting illegal activities.","harmful:illegal-drugs":"Content related to illegal drug use or trade.","harmful:indiscriminate-weapons":"Content related to weapons without context.","harmful:insults":"Content that insults or demeans individuals.","harmful:intellectual-property":"Content violating intellectual property rights.","harmful:misinformation-disinformation":"Spreading false or misleading information.","harmful:non-violent-crime":"Content related to non-violent criminal activities.","harmful:privacy":"Content violating privacy rights.","harmful:profanity":"Content containing profane or inappropriate language.","harmful:radicalization":"Content that promotes radical or extremist views.","harmful:self-harm":"Content that encourages self-harm or suicide.","harmful:sex-crime":"Content related to sexual crimes.","harmful:sexual-content":"Explicit or inappropriate sexual content.","harmful:specialized-advice":"Providing advice in specialized fields without expertise.","harmful:unsafe-practices":"Content promoting unsafe or harmful practices.","harmful:violent-crime":"Content related to violent criminal activities.","prompt-injection":"Malicious inputs designed to manipulate the model's behavior.",competitors:"Competitor mentions and endorsements",contracts:"Enters business or legal commitments without supervision.",hallucination:"Model generating false or misleading information.",hijacking:"Unauthorized or off-topic resource use.",jailbreak:"Bypassing security measures or restrictions.",overreliance:"Model susceptible to relying on an incorrect user assumption or input.",pii:"Exposure or misuse of personally identifiable information.",politics:"Makes political statements."};(a=r||(r={})).Critical="Critical",a.High="High",a.Medium="Medium",a.Low="Low";let v={"excessive-agency":r.Medium,"harmful:child-exploitation":r.Critical,"harmful:copyright-violations":r.Medium,"harmful:cybercrime":r.High,"harmful:graphic-content":r.Medium,"harmful:harassment-bullying":r.High,"harmful:hate":r.Critical,"harmful:illegal-activities":r.High,"harmful:illegal-drugs":r.High,"harmful:indiscriminate-weapons":r.Medium,"harmful:insults":r.Low,"harmful:intellectual-property":r.Medium,"harmful:misinformation-disinformation":r.Medium,"harmful:non-violent-crime":r.Medium,"harmful:privacy":r.High,"harmful:profanity":r.Low,"harmful:radicalization":r.High,"harmful:self-harm":r.Critical,"harmful:sex-crime":r.Critical,"harmful:sexual-content":r.High,"harmful:specialized-advice":r.High,"harmful:unsafe-practices":r.Low,"harmful:violent-crime":r.Critical,"prompt-injection":r.Medium,competitors:r.Low,contracts:r.Medium,hallucination:r.Medium,hijacking:r.High,jailbreak:r.Medium,overreliance:r.Low,pii:r.High,politics:r.Low},x=Object.entries(p).reduce((e,i)=>{let[t,a]=i;return a.forEach(i=>{e[i]=t}),e},{});Object.keys(x);let y={"excessive-agency":"ExcessiveAgency","harmful:child-exploitation":"Child Exploitation","harmful:copyright-violations":"Copyright Violations","harmful:cybercrime":"Cybercrime","harmful:graphic-content":"Graphic Content","harmful:harassment-bullying":"Harassment & Bullying","harmful:hate":"Hate","harmful:illegal-activities":"Illegal Activities","harmful:illegal-drugs":"Illegal Drugs","harmful:indiscriminate-weapons":"Indiscriminate Weapons","harmful:insults":"Insults","harmful:intellectual-property":"Intellectual Property","harmful:misinformation-disinformation":"Misinformation & Disinformation","harmful:non-violent-crime":"Non-Violent Crime","harmful:privacy":"Privacy","harmful:profanity":"Profanity","harmful:radicalization":"Radicalization","harmful:self-harm":"Self-Harm","harmful:sex-crime":"Sex Crimes","harmful:sexual-content":"Sexual Content","harmful:specialized-advice":"Specialized Advice","harmful:unsafe-practices":"Other Unsafe Practices","harmful:violent-crime":"Violent Crimes","prompt-injection":"Harmful/Injection",competitors:"CompetitorEndorsement",contracts:"ContractualCommitment",hallucination:"Hallucination",hijacking:"Hijacking",jailbreak:"Harmful/Iterative",overreliance:"Overreliance",pii:"PIILeak",politics:"PoliticalStatement"},j=Object.entries(y).reduce((e,i)=>{let[t,a]=i;return e[a]=t,e},{}),b={"excessive-agency":"Excessive Agency","prompt-injection":"Prompt Injection",competitors:"Competitor Endorsements",contracts:"Unsupervised Contracts",jailbreak:"Jailbreak",pii:"PII Leaks",politics:"Political Opinions"};t(4309);var w=e=>{let{categoryStats:i}=e,t=[r.Critical,r.High,r.Medium,r.Low],a=t.reduce((e,t)=>(e[t]=Object.keys(i).reduce((e,i)=>v[i]===t?e+1:e,0),e),{});return(0,s.jsx)(m.ZP,{container:!0,spacing:2,mb:4,children:t.map(e=>(0,s.jsx)(m.ZP,{item:!0,xs:12,sm:6,md:3,children:(0,s.jsx)(h.Z,{className:"card-".concat(e.toLowerCase()),children:(0,s.jsxs)(d.Z,{onClick:()=>window.location.hash="#table",children:[(0,s.jsx)(u.Z,{variant:"body2",component:"div",children:e}),(0,s.jsxs)(u.Z,{variant:"h6",color:"text.primary",children:[a[e]," issues"]})]})})},e))})},Z=t(32414),C=t(63362),k=t(18845),R=t(49567),P=t(47908),M=t(85912),N=t(94941),A=t(65969),L=t(70587);t(46553);var I=e=>{let{title:i,subtitle:t,progressValue:a,numTestsPassed:r,numTestsFailed:n,testTypes:l}=e;return(0,s.jsx)(h.Z,{children:(0,s.jsx)(d.Z,{className:"risk-card-container",children:(0,s.jsxs)(m.ZP,{container:!0,spacing:3,children:[(0,s.jsxs)(m.ZP,{item:!0,xs:12,md:6,style:{display:"flex",flexDirection:"column",alignItems:"center",textAlign:"center"},children:[(0,s.jsx)(u.Z,{variant:"h5",className:"risk-card-title",children:i}),(0,s.jsx)(u.Z,{variant:"subtitle1",color:"textSecondary",mb:2,children:t}),(0,s.jsx)(R.Z,{sx:{position:"relative",display:"inline-flex",alignItems:"center",justifyContent:"center",width:100,height:100},children:(0,s.jsx)(L.a,{value:a,max:100,thickness:10,arc:{startAngle:-90,endAngle:90,color:"primary.main"},text:"".concat(Math.round(a),"%"),sx:{width:"100%",height:"100%"}})}),(0,s.jsxs)(u.Z,{variant:"h6",className:"risk-card-issues",children:[n," failed probes"]}),(0,s.jsxs)(u.Z,{variant:"subtitle1",color:"textSecondary",className:"risk-card-tests-passed",children:[r,"/",r+n," passed"]})]}),(0,s.jsx)(m.ZP,{item:!0,xs:6,md:4,children:(0,s.jsx)(P.Z,{dense:!0,children:l.map((e,i)=>(0,s.jsx)(A.Z,{title:g[e.name],placement:"left",arrow:!0,children:(0,s.jsxs)(M.ZP,{className:"risk-card-list-item",onClick:()=>{let i=new URLSearchParams(window.location.search),t=i.get("evalId"),a=y[e.name];window.location.href="/eval/?evalId=".concat(t,"&search=").concat(encodeURIComponent("(var=".concat(a,"|metric=").concat(a,")")))},style:{cursor:"pointer"},children:[(0,s.jsx)(N.Z,{primary:b[e.name]||y[e.name],primaryTypographyProps:{variant:"body2"}}),e.passed?(0,s.jsx)(k.Z,{className:"risk-card-icon-passed"}):(0,s.jsx)(C.Z,{className:"risk-card-icon-failed"})]})},i))})})]})})})};t(32991);var F=e=>{let{categoryStats:i}=e,t=Object.keys(p).map(e=>({name:e,passed:p[e].every(e=>{var t,a;return(null===(t=i[e])||void 0===t?void 0:t.pass)===(null===(a=i[e])||void 0===a?void 0:a.total)})}));return(0,s.jsx)(Z.Z,{spacing:4,children:t.map((e,t)=>{let a=e.name,r=p[a],n=r.reduce((e,t)=>{var a;return e+((null===(a=i[t])||void 0===a?void 0:a.pass)||0)},0),l=r.reduce((e,t)=>{var a;return e+((null===(a=i[t])||void 0===a?void 0:a.total)||0)},0);return(0,s.jsx)(I,{title:e.name,subtitle:f[a],progressValue:n/l*100,numTestsPassed:n,numTestsFailed:l-n,testTypes:r.map(e=>{var t,a;return{name:e,passed:(null===(t=i[e])||void 0===t?void 0:t.pass)===(null===(a=i[e])||void 0===a?void 0:a.total)}})},t)})})},H=t(40982),S=t(37803),z=t(65068),E=t(70417),U=t(21303),D=t(61451),O=t(26485),W=t(38592),B=t(30021),T=t(19708),V=t(28891);t(93091);let _=e=>{let i=[];for(let[t,a]of Object.entries(p))for(let t of a)i.push({pluginName:t,type:y[t]||t,description:g[t]||"",passRate:e[t]?(e[t].pass/e[t].total*100).toFixed(1)+"%":"N/A",passRateWithFilter:e[t]?(e[t].passWithFilter/e[t].total*100).toFixed(1)+"%":"N/A",severity:v[t]||"Unknown"});return i.sort((e,i)=>"N/A"===e.passRate?1:"N/A"===i.passRate?-1:parseFloat(e.passRate)-parseFloat(i.passRate))};var q=e=>{let{evalId:i,categoryStats:t}=e;(0,V.useRouter)();let a=_(t).filter(e=>"N/A"!==e.passRate),[r,l]=n.useState(0),[o,c]=n.useState(10),[m,h]=n.useState("asc"),[d,p]=n.useState("default"),f=e=>{let i=d===e&&"asc"===m;h(i?"desc":"asc"),p(e)};return(0,s.jsxs)(R.Z,{children:[(0,s.jsx)(u.Z,{variant:"h6",gutterBottom:!0,id:"table",children:"Vulnerabilities and Mitigations"}),(0,s.jsxs)(D.Z,{component:S.Z,children:[(0,s.jsxs)(z.Z,{children:[(0,s.jsx)(O.Z,{children:(0,s.jsxs)(B.Z,{children:[(0,s.jsx)(U.Z,{children:"Type"}),(0,s.jsx)(U.Z,{children:"Description"}),(0,s.jsx)(U.Z,{children:(0,s.jsx)(T.Z,{active:"passRate"===d,direction:"passRate"===d?m:"asc",onClick:()=>f("passRate"),children:"Pass rate"})}),(0,s.jsx)(U.Z,{children:(0,s.jsx)(T.Z,{active:"severity"===d,direction:"severity"===d?m:"asc",onClick:()=>f("severity"),children:"Severity"})}),(0,s.jsx)(U.Z,{children:"Actions"})]})}),(0,s.jsx)(E.Z,{children:a.sort((e,i)=>{if("passRate"===d)return"N/A"===e.passRate?1:"N/A"===i.passRate?-1:"asc"===m?parseFloat(e.passRate)-parseFloat(i.passRate):parseFloat(i.passRate)-parseFloat(e.passRate);if("severity"===d){if("N/A"===e.passRate)return 1;if("N/A"===i.passRate)return -1;let t={Critical:4,High:3,Medium:2,Low:1};return"asc"===m?t[e.severity]-t[i.severity]:t[i.severity]-t[e.severity]}{let t={Critical:4,High:3,Medium:2,Low:1};return e.severity===i.severity?parseFloat(e.passRate)-parseFloat(i.passRate):t[i.severity]-t[e.severity]}}).slice(r*o,r*o+o).map((e,i)=>{let t="";if("N/A"!==e.passRate){let i=parseFloat(e.passRate);t=i>=75?"pass-high":i>=50?"pass-medium":"pass-low"}return(0,s.jsxs)(B.Z,{children:[(0,s.jsx)(U.Z,{children:b[e.pluginName]||e.type}),(0,s.jsx)(U.Z,{children:e.description}),(0,s.jsxs)(U.Z,{className:t,children:[e.passRate," (",e.passRateWithFilter!==e.passRate?(0,s.jsx)(s.Fragment,{children:(0,s.jsxs)("strong",{children:[e.passRateWithFilter," with mitigation"]})}):e.passRateWithFilter,")"]}),(0,s.jsx)(U.Z,{className:"vuln-".concat(e.severity.toLowerCase()),children:e.severity}),(0,s.jsxs)(U.Z,{style:{minWidth:270},children:[(0,s.jsx)(H.Z,{variant:"contained",size:"small",onClick:()=>{let i=new URLSearchParams(window.location.search),t=i.get("evalId");window.location.href="/eval/?evalId=".concat(t,"&search=").concat(encodeURIComponent("(var=".concat(e.type,"|metric=").concat(e.type,")")))},children:"View logs"}),(0,s.jsx)(A.Z,{title:"Temporarily disabled while in beta, click to contact us to enable",children:(0,s.jsx)(H.Z,{variant:"contained",size:"small",color:"inherit",style:{marginLeft:8},onClick:()=>{window.location.href="mailto:inquiries@promptfoo.dev?subject=Promptfoo%20automatic%20vulnerability%20mitigation&body=Hello%20Promptfoo%20Team,%0D%0A%0D%0AI%20am%20interested%20in%20learning%20more%20about%20the%20automatic%20vulnerability%20mitigation%20beta.%20Please%20provide%20me%20with%20more%20details.%0D%0A%0D%0A"},children:"Apply mitigation"})})]})]},i)})})]}),a.length>o&&(0,s.jsx)(W.Z,{rowsPerPageOptions:[10,25,50],component:"div",count:a.length,rowsPerPage:o,page:r,onPageChange:(e,i)=>{l(i)},onRowsPerPageChange:e=>{c(parseInt(e.target.value,10)),l(0)}})]})]})};t(67667);var G=()=>{let[e,i]=n.useState(null),[t,a]=n.useState(null);if(n.useEffect(()=>{let e=async e=>{let i=await fetch("".concat(await (0,l.b)(),"/api/results/").concat(e),{cache:"no-store"}),t=await i.json();a(t.data)},t=new URLSearchParams(window.location.search);if(!t)return;let r=t.get("evalId");r&&(i(r),e(r))},[]),!t||!e)return(0,s.jsx)("div",{children:"Loading..."});let r=t.results.table.head.prompts[0],h=t.results.table.body,d=t.results.results.reduce((e,i)=>{var t,a,r,s;let n=i.vars.harmCategory,l=(null===(a=i.gradingResult)||void 0===a?void 0:null===(t=a.componentResults)||void 0===t?void 0:t.map(e=>{var i;return null===(i=e.assertion)||void 0===i?void 0:i.metric}))||[],o=[n,...l].filter(e=>e);for(let t of o){let a=j[t];if(!a){console.log("Unknown harm category:",t);break}let n=i.success;e[a]=e[a]||{pass:0,total:0,passWithFilter:0},e[a].total++,n?(e[a].pass++,e[a].passWithFilter++):(null===(s=i.gradingResult)||void 0===s?void 0:null===(r=s.componentResults)||void 0===r?void 0:r.some(e=>{var i;let t=(null===(i=e.assertion)||void 0===i?void 0:i.type)==="moderation",a=!e.pass;return t&&a}))&&e[a].passWithFilter++}return e},{});return(0,s.jsx)(c.Z,{children:(0,s.jsxs)(m.ZP,{container:!0,direction:"column",spacing:1,pt:6,pb:8,children:[(0,s.jsxs)(m.ZP,{item:!0,className:"report-header",children:[(0,s.jsxs)(u.Z,{variant:"h4",children:[(0,s.jsx)("strong",{children:"LLM Risk Report"}),t.config.description&&": ".concat(t.config.description)]}),(0,s.jsx)(u.Z,{variant:"subtitle1",mb:2,children:new Date(t.createdAt).toLocaleDateString("en-US",{year:"numeric",month:"long",day:"numeric"})}),(0,s.jsxs)(u.Z,{variant:"body1",gutterBottom:!0,className:"report-details",children:[(0,s.jsx)(o.Z,{size:"small",label:(0,s.jsxs)(s.Fragment,{children:[(0,s.jsx)("strong",{children:"Model:"})," ",r.provider]})}),(0,s.jsx)(o.Z,{size:"small",label:(0,s.jsxs)(s.Fragment,{children:[(0,s.jsx)("strong",{children:"Dataset:"})," ",h.length," probes"]})}),(0,s.jsx)(o.Z,{size:"small",label:(0,s.jsxs)(s.Fragment,{children:[(0,s.jsx)("strong",{children:"Prompt:"}),' "',r.raw.length>20?"".concat(r.raw.substring(0,20),"..."):r.raw,'"']})})]})]}),(0,s.jsx)(m.ZP,{item:!0,children:(0,s.jsx)(w,{categoryStats:d})}),(0,s.jsx)(m.ZP,{item:!0,children:(0,s.jsx)(F,{categoryStats:d})}),(0,s.jsx)(m.ZP,{item:!0,children:(0,s.jsx)(q,{evalId:e,categoryStats:d})})]})})}},4309:function(){},67667:function(){},46553:function(){},32991:function(){},93091:function(){}},function(e){e.O(0,[540,215,620,770,304,995,378,915,470,730,744],function(){return e(e.s=25902)}),_N_E=e.O()}]);
@@ -0,0 +1 @@
1
+ (self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[413],{76178:function(e,t,n){Promise.resolve().then(n.bind(n,83013))},83013:function(e,t,n){"use strict";n.r(t),n.d(t,{default:function(){return ed}});var a=n(24004),i=n(14978),r=n(69179),l=n(47887),s=n(49567),o=n(40982),c=n(11520),d=n(29794),p=n(22701),u=n(51956),h=n(8541),x=n(82669),m=n(32414),f=n(63147),g=n(75307),j=n(63346),v=n(89235),Z=n(4078),b=n(67339),y=()=>{let{env:e,setEnv:t}=(0,l.o)(),[n,r]=(0,i.useState)(!1),[s,c]=(0,i.useState)(e),h=()=>{r(!1)};return(0,a.jsxs)(a.Fragment,{children:[(0,a.jsx)(o.Z,{variant:"outlined",startIcon:(0,a.jsx)(g.Z,{}),onClick:()=>{r(!0)},children:"API keys"}),(0,a.jsxs)(d.Z,{open:n,onClose:h,fullWidth:!0,maxWidth:"md",children:[(0,a.jsx)(x.Z,{children:"Provider settings"}),(0,a.jsxs)(u.Z,{children:[(0,a.jsxs)(j.Z,{defaultExpanded:!0,children:[(0,a.jsx)(Z.Z,{children:"OpenAI"}),(0,a.jsxs)(v.Z,{children:[(0,a.jsx)(b.Z,{label:"OpenAI API key",fullWidth:!0,margin:"normal",value:s.OPENAI_API_KEY,onChange:e=>c({...s,OPENAI_API_KEY:e.target.value})}),(0,a.jsx)(b.Z,{label:"OpenAI API host",fullWidth:!0,margin:"normal",value:s.OPENAI_API_HOST,onChange:e=>c({...s,OPENAI_API_HOST:e.target.value})}),(0,a.jsx)(b.Z,{label:"OpenAI organization",fullWidth:!0,margin:"normal",value:s.OPENAI_ORGANIZATION,onChange:e=>c({...s,OPENAI_ORGANIZATION:e.target.value})})]})]}),(0,a.jsxs)(j.Z,{children:[(0,a.jsx)(Z.Z,{children:"Azure"}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Azure API key",fullWidth:!0,margin:"normal",value:s.AZURE_OPENAI_API_KEY,onChange:e=>c({...s,AZURE_OPENAI_API_KEY:e.target.value})})})]}),(0,a.jsxs)(j.Z,{children:[(0,a.jsx)(Z.Z,{children:"Amazon Bedrock"}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Bedrock Region",fullWidth:!0,margin:"normal",value:s.AWS_BEDROCK_REGION,onChange:e=>c({...s,AWS_BEDROCK_REGION:e.target.value})})})]}),(0,a.jsxs)(j.Z,{children:[(0,a.jsx)(Z.Z,{children:"Anthropic"}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Anthropic API key",fullWidth:!0,margin:"normal",value:s.ANTHROPIC_API_KEY,onChange:e=>c({...s,ANTHROPIC_API_KEY:e.target.value})})})]}),(0,a.jsxs)(j.Z,{children:[(0,a.jsx)(Z.Z,{children:"Google Vertex AI"}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Vertex API Key",fullWidth:!0,margin:"normal",value:s.VERTEX_API_KEY,onChange:e=>c({...s,VERTEX_API_KEY:e.target.value})})}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Vertex Project ID",fullWidth:!0,margin:"normal",value:s.VERTEX_PROJECT_ID,onChange:e=>c({...s,VERTEX_PROJECT_ID:e.target.value})})}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Vertex Region",fullWidth:!0,margin:"normal",value:s.VERTEX_REGION,onChange:e=>c({...s,VERTEX_REGION:e.target.value})})})]}),(0,a.jsxs)(j.Z,{children:[(0,a.jsx)(Z.Z,{children:"Replicate"}),(0,a.jsx)(v.Z,{children:(0,a.jsx)(b.Z,{label:"Replicate API key",fullWidth:!0,margin:"normal",value:s.REPLICATE_API_KEY,onChange:e=>c({...s,REPLICATE_API_KEY:e.target.value})})})]})]}),(0,a.jsxs)(p.Z,{children:[(0,a.jsx)(o.Z,{onClick:h,color:"primary",children:"Cancel"}),(0,a.jsx)(o.Z,{onClick:()=>{t(s),h()},color:"primary",variant:"contained",children:"Save"})]})]})]})},_=n(35185),C=n(5592),k=n(14931),A=n(2185),E=n(77656),I=n(65068),O=n(70417),P=n(21303),S=n(61451),T=n(30021),w=n(65969),R=e=>{let{open:t,prompt:n,index:r,onAdd:l,onCancel:s}=e,[c,h]=i.useState(n),m=i.useRef(null);i.useEffect(()=>{h(n)},[n]);let f=e=>{l(c),h(""),e?s():m.current&&m.current.focus()};return(0,a.jsxs)(d.Z,{open:t,onClose:s,fullWidth:!0,maxWidth:"md",children:[(0,a.jsx)(x.Z,{children:"Edit Prompt ".concat(r+1)}),(0,a.jsx)(u.Z,{children:(0,a.jsx)(b.Z,{value:c,onChange:e=>h(e.target.value),fullWidth:!0,margin:"normal",multiline:!0,placeholder:"The quick brown {{animal1}} jumps over the lazy {{animal2}}.",helperText:"Tip: use the {{varname}} syntax to add variables to your prompt.",inputRef:m})}),(0,a.jsxs)(p.Z,{children:[(0,a.jsx)(o.Z,{onClick:f.bind(null,!0),color:"primary",variant:"contained",disabled:!c.length,children:"Add"}),(0,a.jsx)(o.Z,{onClick:f.bind(null,!1),color:"primary",variant:"contained",disabled:!c.length,children:"Add Another"}),(0,a.jsx)(o.Z,{onClick:s,color:"secondary",children:"Cancel"})]})]})};n(94499);var W=()=>{let[e,t]=(0,i.useState)(!1),[n,r]=(0,i.useState)(null),{prompts:s,setPrompts:c}=(0,l.o)(),d=(0,i.useRef)(null);(0,i.useEffect)(()=>{null!==n&&n>0&&d.current&&d.current.focus()},[n]);let p=e=>{r(e),t(!0)},u=(e,t)=>{e.stopPropagation();let n=s[t];c([...s,n])},h=(e,t)=>{c(s.map((n,a)=>a===e?t:n))},x=(e,t)=>{e.stopPropagation(),confirm("Are you sure you want to remove this prompt?")&&c(s.filter((e,n)=>n!==t))};return(0,a.jsxs)("div",{children:[(0,a.jsxs)(m.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,a.jsx)(f.Z,{variant:"h5",children:"Prompts"}),(0,a.jsxs)("div",{children:[(0,a.jsx)("label",{htmlFor:"file-input-add-prompt",children:(0,a.jsx)(w.Z,{title:"Upload prompt from file",children:(0,a.jsxs)("span",{children:[(0,a.jsx)(E.Z,{component:"span",children:(0,a.jsx)(A.Z,{})}),(0,a.jsx)("input",{id:"file-input-add-prompt",type:"file",accept:".txt,.md",onChange:e=>{var t;e.stopPropagation(),e.preventDefault();let n=null===(t=e.target.files)||void 0===t?void 0:t[0];if(n){let e=new FileReader;e.onload=e=>{var t,n;let a=null===(n=e.target)||void 0===n?void 0:null===(t=n.result)||void 0===t?void 0:t.toString();a&&c([...s,a])},e.readAsText(n)}},style:{display:"none"}})]})})}),(0,a.jsx)(o.Z,{color:"primary",onClick:()=>{t(!0)},variant:"contained",children:"Add Prompt"})]})]}),(0,a.jsx)(S.Z,{children:(0,a.jsx)(I.Z,{children:(0,a.jsx)(O.Z,{children:0===s.length?(0,a.jsx)(T.Z,{children:(0,a.jsx)(P.Z,{colSpan:2,align:"center",children:"No prompts added yet."})}):s.map((e,t)=>(0,a.jsxs)(T.Z,{sx:{"&:hover":{backgroundColor:"rgba(0, 0, 0, 0.04)",cursor:"pointer"}},onClick:()=>p(t),children:[(0,a.jsx)(P.Z,{children:(0,a.jsxs)(f.Z,{variant:"body2",children:["Prompt #".concat(t+1,": "),(e.length>250?e.slice(0,250)+" ...":e).split(/({{\w+}})/g).map((e,t)=>/{{\s*(\w+)\s*}}/g.test(e)?(0,a.jsx)("span",{className:"prompt-var-highlight",children:e},t):e)]})}),(0,a.jsxs)(P.Z,{align:"right",sx:{minWidth:150},children:[(0,a.jsx)(E.Z,{onClick:()=>p(t),size:"small",children:(0,a.jsx)(k.Z,{})}),(0,a.jsx)(E.Z,{onClick:e=>u(e,t),size:"small",children:(0,a.jsx)(_.Z,{})}),(0,a.jsx)(E.Z,{onClick:e=>x(e,t),size:"small",children:(0,a.jsx)(C.Z,{})})]})]},t))})})}),(0,a.jsx)(R,{open:e,prompt:null!==n?s[n]:"",index:null!==n?n:0,onAdd:e=>{null!==n?h(n,e):c([...s,e]),r(null)},onCancel:()=>{r(null),t(!1)}})]})},N=n(68873),z=n(885),F=e=>{let{onChange:t,...n}=e,[r,l]=i.useState(""),[s,o]=i.useState(!1);return(0,a.jsx)(b.Z,{...n,error:s,helperText:s?"Invalid JSON":"",value:r,onChange:e=>{let n=e.target.value;try{let e=JSON.parse(n);l(n),o(!1),t&&t(e)}catch(e){l(n),o(!0)}}})},V=e=>{let{open:t,providerId:n,config:r,onClose:l,onSave:c}=e,[h,m]=i.useState(r);return i.useEffect(()=>{m(r)},[r]),(0,a.jsxs)(d.Z,{open:t,onClose:l,fullWidth:!0,maxWidth:"md",children:[(0,a.jsxs)(x.Z,{children:["Edit ",n.length>50?n.slice(0,50)+"...":n]}),(0,a.jsx)(u.Z,{children:Object.keys(h).map(e=>{let t;let n=h[e];return"number"==typeof n||"boolean"==typeof n||"string"==typeof n?(t="number"==typeof n?t=>m({...h,[e]:parseFloat(t.target.value)}):"boolean"==typeof n?t=>m({...h,[e]:"true"===t.target.value}):t=>{let n=t.target.value.trim();if(n.startsWith("{")||n.startsWith("["))try{m({...h,[e]:JSON.parse(n)})}catch(t){m({...h,[e]:n})}else"null"===n?m({...h,[e]:null}):"undefined"===n?m({...h,[e]:void 0}):m({...h,[e]:n})},(0,a.jsx)(s.Z,{my:2,children:(0,a.jsx)(b.Z,{label:e,value:n,onChange:t,fullWidth:!0,InputLabelProps:{shrink:!0},type:"number"==typeof n?"number":"text"})},e)):(0,a.jsx)(s.Z,{my:2,children:(0,a.jsx)(F,{label:e,defaultValue:JSON.stringify(n),onChange:t=>{m({...h,[e]:t})},fullWidth:!0,multiline:!0,minRows:2,InputLabelProps:{shrink:!0}})},e)})}),(0,a.jsxs)(p.Z,{children:[(0,a.jsx)(o.Z,{onClick:l,children:"Cancel"}),(0,a.jsx)(o.Z,{onClick:()=>{c(h)},children:"Save"})]})]})};let L=[].concat(["replicate:replicate/flan-t5-small:69716ad8c34274043bf4a135b7315c7c569ec931d8f23d6826e249e1c142a264"].map(e=>({id:e,config:{temperature:.5,max_length:1024,repetition_penality:1}}))).concat(["replicate:replicate/codellama-7b-instruct:0103579e86fc75ba0d65912890fa19ef03c84a68554635319accf2e0ba93d3ae","replicate:replicate/codellama-13b-instruct:da5676342de1a5a335b848383af297f592b816b950a43d251a0a9edd0113604b","replicate:replicate/llama-2-70b-chat:2796ee9483c3fd7aa2e171d38f4ca12251a30609463dcfd4cd76703f22e96cdf"].map(e=>({id:e,config:{system_prompt:"",temperature:.75,top_p:.9,top_k:50,max_new_tokens:128,min_new_tokens:-1}}))).concat(["replicate:replicate/codellama-7b:6880b103613a9cd23950c5fd6c140197e519905bd0dd00e448c4858bdd06090a","replicate:replicate/codellama-13b-python:09b87c02dfa403e0c3289166dece62286b3bce49bae39a9c9204713cf94b8b7d","replicate:replicate/codellama-13b:1c914d844307b0588599b8393480a3ba917b660c7e9dfae681542b5325f228db","replicate:replicate/codellama-34b-python:9048743d22a7b19cd0abb018066809ea6af4f2b4717bef9aad3c5ae21ceac00d","replicate:replicate/codellama-34b:0666717e5ead8557dff55ee8f11924b5c0309f5f1ca52f64bb8eec405fdb38a7"].map(e=>({id:e,config:{temperature:.75,top_p:.9,top_k:50,max_new_tokens:128,min_new_tokens:-1}}))).concat(["replicate:a16z-infra/llama-2-7b-chat:7b0bfc9aff140d5b75bacbed23e91fd3c34b01a1e958d32132de6e0a19796e2c","replicate:a16z-infra/llama-2-13b-chat:2a7f981751ec7fdf87b5b91ad4db53683a98082e9ff7bfd12c8cd5ea85980a52"].map(e=>({id:e,config:{temperature:.95,top_p:.95,top_k:250,max_new_tokens:500,min_new_tokens:-1,repetition_penalty:1,system_prompt:""}}))).concat(["replicate:mistralai/mistral-7b-v0.1","replicate:mistralai/mistral-7b-instruct-v0.2","replicate:mistralai/mixtral-8x7b-instruct-v0.1"].map(e=>({id:e,config:{temperature:.7,top_p:.9,top_k:-1,max_new_tokens:128,min_new_tokens:-1,repetition_penalty:1.15,prompt_template:"{prompt}"}}))).concat(["anthropic:claude-1","anthropic:claude-1-100k","anthropic:claude-instant-1","anthropic:claude-instant-1-100k"].map(e=>({id:e,config:{max_tokens_to_sample:256,temperature:.5}}))).concat(["anthropic:messages:claude-instant-1.2","anthropic:messages:claude-2.0","anthropic:messages:claude-2.1","anthropic:messages:claude-3-haiku-20240307","anthropic:messages:claude-3-sonnet-20240229","anthropic:messages:claude-3-opus-20240229"].map(e=>({id:e,config:{max_tokens:1024,temperature:.5}}))).concat(["bedrock:anthropic.claude-instant-v1","bedrock:anthropic.claude-v1","bedrock:anthropic.claude-v2"].map(e=>({id:e,config:{max_tokens_to_sample:256,temperature:.5}}))).concat(["openai:gpt-3.5-turbo","openai:gpt-3.5-turbo-0301","openai:gpt-3.5-turbo-0613","openai:gpt-3.5-turbo-16k","openai:gpt-3.5-turbo-16k-0613","openai:gpt-4","openai:gpt-4-0314","openai:gpt-4-0613","openai:gpt-4-32k","openai:gpt-4-32k-0314"].map(e=>({id:e,config:{organization:"",temperature:.5,max_tokens:1024,top_p:1,frequency_penalty:0,presence_penalty:0,function_call:void 0,functions:void 0,stop:void 0}}))).concat(["azureopenai:gpt-3.5-turbo","azureopenai:gpt-3.5-turbo-0301","azureopenai:gpt-3.5-turbo-0613","azureopenai:gpt-3.5-turbo-16k","azureopenai:gpt-3.5-turbo-16k-0613","azureopenai:gpt-4","azureopenai:gpt-4-0314","azureopenai:gpt-4-0613","azureopenai:gpt-4-32k","azureopenai:gpt-4-32k-0314"].map(e=>({id:e,config:{temperature:.5,max_tokens:1024,top_p:1,frequency_penalty:0,presence_penalty:0,function_call:void 0,functions:void 0,stop:void 0}}))).concat(["vertex:chat-bison@001","vertex:chat-bison","vertex:chat-bison-32k","vertex:chat-bison-32k@001"].map(e=>({id:e,config:{context:void 0,examples:void 0,temperature:0,maxOutputTokens:1024,topP:.95,topK:40,safetySettings:void 0,stopSequence:void 0}}))).sort((e,t)=>e.id.localeCompare(t.id)),U={anthropic:"Anthropic",bedrock:"Amazon Web Services",azureopenai:"Azure",openai:"OpenAI",replicate:"Replicate"};var q=e=>{let{providers:t,onChange:n}=e,[r,l]=i.useState(null),o=e=>"string"==typeof e?e:e.id||"Unknown provider",c=(e,t)=>"string"==typeof e?e:e.id||t,d=e=>{"string"==typeof e?alert("Cannot edit custom providers"):e.config?l(e):alert("There is no config for this provider")};return(0,a.jsxs)(s.Z,{mt:2,children:[(0,a.jsx)(N.Z,{multiple:!0,freeSolo:!0,options:L,value:t,groupBy:e=>(function(e){if(!e)return"Other";let t=e.split(":")[0];return U[t]||t})(e.id),onChange:(e,t)=>{n(t.map(e=>"string"==typeof e?{id:e}:e))},getOptionLabel:e=>{if(!e)return"";let t="";"string"==typeof e&&(t=e),e.id&&"string"==typeof e.id&&(t=e.id);let n=t.split(":");return n.length>1?n.length>2&&"anthropic"===n[0]?n[2]:n[1]:"Unknown provider"},renderTags:(e,t)=>e.map((e,n)=>{let a=o(e),r=c(e,n);return(0,i.createElement)(z.Z,{variant:"outlined",label:a,...t({index:n}),key:r,onClick:()=>d(e)})}),renderInput:e=>(0,a.jsx)(b.Z,{...e,variant:"outlined",placeholder:"Select LLM providers",helperText:t.length>0?"Click a provider to configure its settings.":null})}),r&&r.id&&(0,a.jsx)(V,{open:!!r,providerId:r.id,config:r.config,onClose:()=>l(null),onSave:e=>{if(r){let a=t.map(t=>t.id===r.id?{...t,config:e}:t);n(a),l(null)}}})]})},B=n(52428),K=n(11615),D=n(28891),Y=()=>{let e=(0,D.useRouter)(),{env:t,description:n,providers:r,prompts:s,testCases:c,defaultTest:d,evaluateOptions:p}=(0,l.o)(),[u,h]=(0,i.useState)(!1),[x,m]=(0,i.useState)(0),f=async()=>{h(!0);try{let a=await fetch("".concat(B.eA,"/api/eval/job/"),{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({env:t,description:n,providers:r,prompts:s,tests:c,defaultTest:d,evaluateOptions:p})});if(!a.ok)throw Error("HTTP error! status: ".concat(a.status));let i=await a.json(),l=setInterval(async()=>{let t=await fetch("".concat(B.eA,"/api/eval/job/").concat(i.id,"/"));if(!t.ok)throw clearInterval(l),Error("HTTP error! status: ".concat(t.status));let n=await t.json();if("complete"===n.status)clearInterval(l),h(!1),B.Ox?e.push("/eval/remote:".concat(encodeURIComponent(i.id))):e.push("/eval");else if("failed"===n.status)throw clearInterval(l),h(!1),Error("Job failed");else{let e=0===n.total?0:Math.round(n.progress/n.total*100);m(e)}},1e3)}catch(e){console.error(e),h(!1),alert("An error occurred: ".concat(e.message))}};return(0,a.jsx)(o.Z,{variant:"contained",color:"primary",onClick:f,disabled:u,children:u?(0,a.jsxs)(a.Fragment,{children:[(0,a.jsx)(K.Z,{size:24,sx:{marginRight:2}}),x.toFixed(0),"% complete"]}):"Run Evaluation"})},J=n(15320),X=n(26485);let G=["equals","contains","icontains","contains-all","contains-any","starts-with","regex","is-json","contains-json","is-sql","contains-sql","similar","llm-rubric","model-graded-closedqa","factuality","webhook","rouge-n","rouge-s","rouge-l","not-equals","not-contains","not-icontains","not-contains-all","not-contains-any","not-starts-with","not-regex","not-is-json","not-contains-json","not-similar","not-webhook","not-rouge-n","not-rouge-s","not-rouge-l","is-valid-openai-function-call","is-valid-openai-tools-call","latency","perplexity","perplexity-score","cost","answer-relevance","context-faithfulness","context-recall","context-relevance","select-best","moderation"];var H=e=>{let{onAdd:t,initialValues:n}=e,[r,l]=(0,i.useState)(n||[]),c=e=>{let n=r.filter((t,n)=>n!==e);l(n),t(n)};return(0,a.jsxs)(a.Fragment,{children:[(0,a.jsx)(f.Z,{variant:"h6",children:"Asserts"}),(0,a.jsx)(s.Z,{my:r.length>0?2:0,children:(0,a.jsx)(m.Z,{direction:"column",spacing:2,children:r.map((e,n)=>(0,a.jsxs)(m.Z,{direction:"row",spacing:2,alignItems:"center",children:[(0,a.jsx)(N.Z,{value:e.type,options:G,sx:{minWidth:200},onChange:(e,a)=>{let i=r.map((e,t)=>t===n?{...e,type:a}:e);l(i),t(i)},renderInput:e=>(0,a.jsx)(b.Z,{...e,label:"Type"})}),(0,a.jsx)(b.Z,{label:"Value",value:e.value,fullWidth:!0,onChange:e=>{let a=e.target.value,i=r.map((e,t)=>t===n?{...e,value:a}:e);l(i),t(i)}}),(0,a.jsx)(E.Z,{onClick:()=>c(n),size:"small",children:(0,a.jsx)(C.Z,{})})]},n))})}),(0,a.jsx)(o.Z,{color:"primary",onClick:()=>{let e=[...r,{type:"equals",value:""}];l(e),t(e)},children:"Add Assert"})]})},M=e=>{let{onAdd:t,varsList:n,initialValues:r}=e,[l,o]=i.useState(r||{});return(0,i.useEffect)(()=>{let e={};n.forEach(t=>{e[t]=(null==r?void 0:r[t])||""}),o(e)},[n,r]),(0,a.jsxs)(s.Z,{my:2,children:[(0,a.jsx)(f.Z,{variant:"h6",mb:2,children:"Vars"}),n.length>0?(0,a.jsx)(m.Z,{direction:"row",spacing:2,alignItems:"center",children:Object.keys(l).map((e,n)=>(0,a.jsx)(m.Z,{direction:"row",spacing:2,alignItems:"center",children:(0,a.jsx)(b.Z,{placeholder:e,label:e,value:l[e],fullWidth:!0,onChange:n=>{let a=n.target.value,i={...l,[e]:a};o(i),t(i)}})},n))}):(0,a.jsxs)(f.Z,{variant:"subtitle1",gutterBottom:!0,children:["Add variables to your prompt using the ","{{varname}}"," syntax."]})]})},$=e=>{let{open:t,onAdd:n,varsList:r,initialValues:l,onCancel:c}=e,[h,m]=(0,i.useState)((null==l?void 0:l.description)||""),[f,g]=(0,i.useState)((null==l?void 0:l.vars)||{}),[j,v]=(0,i.useState)((null==l?void 0:l.assert)||[]),[Z,b]=(0,i.useState)(0);i.useEffect(()=>{l?(m(l.description||""),g(l.vars||{}),v(l.assert||[])):(m(""),g({}),v([]))},[l]);let y=e=>{n({description:h,vars:f,assert:j},e),e&&c(),m(""),g({}),v([]),b(e=>e+1)};return(0,a.jsxs)(d.Z,{open:t,onClose:c,fullWidth:!0,maxWidth:"md",children:[(0,a.jsx)(x.Z,{children:l?"Edit Test Case":"Add Test Case"}),(0,a.jsx)(u.Z,{children:(0,a.jsxs)(s.Z,{children:[(0,a.jsx)(M,{onAdd:e=>g(e),varsList:r,initialValues:null==l?void 0:l.vars}),(0,a.jsx)(H,{onAdd:e=>v(e),initialValues:((null==l?void 0:l.assert)||[]).filter(e=>"assert-set"!==e.type)||[]},Z)]})}),(0,a.jsxs)(p.Z,{children:[(0,a.jsx)(o.Z,{onClick:y.bind(void 0,!0),color:"primary",variant:"contained",children:l?"Update Test Case":"Add Test Case"}),!l&&(0,a.jsx)(o.Z,{onClick:y.bind(void 0,!1),color:"primary",variant:"contained",children:"Add Another"}),(0,a.jsx)(o.Z,{onClick:c,color:"secondary",children:"Cancel"})]})]})},Q=e=>{let{varsList:t}=e,{testCases:r,setTestCases:s}=(0,l.o)(),[c,d]=i.useState(null),[p,u]=i.useState(!1),h=(e,t)=>{e.stopPropagation(),confirm("Are you sure you want to delete this test case?")&&s(r.filter((e,n)=>n!==t))},x=(e,t)=>{e.stopPropagation();let n=JSON.parse(JSON.stringify(r[t]));s([...r,n])};return(0,a.jsxs)(a.Fragment,{children:[(0,a.jsxs)(m.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,a.jsx)(f.Z,{variant:"h5",children:"Test Cases"}),(0,a.jsxs)("div",{children:[(0,a.jsx)("label",{htmlFor:"file-input-add-test-case",children:(0,a.jsx)(w.Z,{title:"Upload test cases from csv",children:(0,a.jsxs)("span",{children:[(0,a.jsx)(E.Z,{component:"span",children:(0,a.jsx)(A.Z,{})}),(0,a.jsx)("input",{id:"file-input-add-test-case",type:"file",accept:".csv",onChange:e=>{var t;e.stopPropagation(),e.preventDefault();let a=null===(t=e.target.files)||void 0===t?void 0:t[0];if(a){let e=new FileReader;e.onload=async e=>{var t,a;let i=null===(a=e.target)||void 0===a?void 0:null===(t=a.result)||void 0===t?void 0:t.toString();if(i){let{parse:e}=await Promise.all([n.e(631),n.e(887)]).then(n.bind(n,83887)),t=e(i,{columns:!0});s([...r,...t.map(e=>(0,J.It)(e))])}},e.readAsText(a)}},style:{display:"none"}})]})})}),(0,a.jsx)(o.Z,{color:"primary",onClick:()=>u(!0),variant:"contained",children:"Add Test Case"})]})]}),(0,a.jsx)(S.Z,{children:(0,a.jsxs)(I.Z,{children:[(0,a.jsx)(X.Z,{children:(0,a.jsxs)(T.Z,{children:[(0,a.jsx)(P.Z,{children:"Description"}),(0,a.jsx)(P.Z,{children:"Assertions"}),(0,a.jsx)(P.Z,{children:"Variables"}),(0,a.jsx)(P.Z,{align:"right"})]})}),(0,a.jsx)(O.Z,{children:0===r.length?(0,a.jsx)(T.Z,{children:(0,a.jsx)(P.Z,{colSpan:4,align:"center",children:"No test cases added yet."})}):r.map((e,t)=>{var n;return(0,a.jsxs)(T.Z,{sx:{"&:hover":{backgroundColor:"rgba(0, 0, 0, 0.04)",cursor:"pointer"}},onClick:()=>{d(t),u(!0)},children:[(0,a.jsx)(P.Z,{children:(0,a.jsx)(f.Z,{variant:"body2",children:e.description||"Test Case #".concat(t+1)})}),(0,a.jsxs)(P.Z,{children:[(null===(n=e.assert)||void 0===n?void 0:n.length)||0," assertions"]}),(0,a.jsx)(P.Z,{children:Object.entries(e.vars||{}).map(e=>{let[t,n]=e;return t+"="+n}).join(", ")}),(0,a.jsxs)(P.Z,{align:"right",sx:{minWidth:150},children:[(0,a.jsx)(E.Z,{onClick:()=>{d(t),u(!0)},size:"small",children:(0,a.jsx)(k.Z,{})}),(0,a.jsx)(E.Z,{onClick:e=>x(e,t),size:"small",children:(0,a.jsx)(_.Z,{})}),(0,a.jsx)(E.Z,{onClick:e=>h(e,t),size:"small",children:(0,a.jsx)(C.Z,{})})]})]},t)})})]})}),(0,a.jsx)($,{open:p,onAdd:(e,t)=>{if(null===c)s([...r,e]);else{let t=r.map((t,n)=>n===c?e:t);s(t),d(null)}t&&u(!1)},varsList:t,initialValues:null!==c?r[c]:void 0,onCancel:()=>{d(null),u(!1)}})]})},ee=n(71911),et=n.n(ee),en=n(93522),ea=n(34235),ei=n(38640),er=n.n(ei),el=n(29270);n(54492),n(55096),n(93486);var es=()=>{let{env:e,setEnv:t,description:n,setDescription:r,providers:c,setProviders:d,prompts:p,setPrompts:u,testCases:h,setTestCases:x,defaultTest:m,setDefaultTest:g,evaluateOptions:j,setEvaluateOptions:v}=(0,l.o)(),[Z,b]=i.useState(""),[y,_]=i.useState(!0);i.useEffect(()=>{b(ea.default.dump({env:e,description:n,providers:c,prompts:p,tests:h,defaultTest:m,evaluateOptions:j}))},[e,n,c,p,h,m,j]);let C=e=>{t(e.env||{}),r(e.description||""),d(e.providers||[]),u(e.prompts||[]),x(e.tests||[]),g(e.defaultTest||{}),v(e.evaluateOptions||{})};return(0,a.jsxs)(s.Z,{mt:4,children:[(0,a.jsx)(f.Z,{variant:"h5",gutterBottom:!0,children:"Configuration"}),(0,a.jsxs)(f.Z,{variant:"body1",gutterBottom:!0,children:["This is the YAML config that defines the evaluation and is processed by promptfoo. See"," ",(0,a.jsx)(er(),{target:"_blank",href:"https://promptfoo.dev/docs/configuration/guide",children:"configuration docs"})," ","to learn more."]}),(0,a.jsx)(o.Z,{variant:"text",color:"primary",startIcon:y?(0,a.jsx)(k.Z,{}):(0,a.jsx)(en.Z,{}),onClick:()=>{if(!y)try{let e=ea.default.load(Z,{json:!0});C(e)}catch(e){}_(!y)},children:y?"Edit YAML":"Save"}),(0,a.jsx)(et(),{autoCapitalize:"off",value:Z,onValueChange:e=>{y||b(e)},highlight:e=>(0,el.highlight)(e,el.languages.yaml),padding:10,style:{fontFamily:'"Fira code", "Fira Mono", monospace',fontSize:14},disabled:y,className:y?"":"glowing-border"})]})};n(7517);var eo=n(77580);function ec(e){let{error:t,resetErrorBoundary:n}=e;return(0,a.jsxs)("div",{role:"alert",children:[(0,a.jsx)("p",{children:"Something went wrong:"}),(0,a.jsx)("pre",{children:t.message}),(0,a.jsx)("button",{onClick:n,children:"Try again"})]})}var ed=()=>{let[e,t]=(0,i.useState)(!1),{description:n,setDescription:g,providers:j,setProviders:v,prompts:Z,setPrompts:b,testCases:_,setTestCases:C}=(0,l.o)();if((0,i.useEffect)(()=>{l.o.persist.rehydrate()},[]),eo.env.NEXT_PUBLIC_NO_BROWSING)return null;let k=(e=>{let t=/{{\s*(\w+)\s*}}/g,n=new Set;return e.forEach(e=>{let a;for(;null!==(a=t.exec(e));)n.add(a[1])}),Array.from(n)})(Z);return(0,a.jsxs)(c.Z,{maxWidth:"lg",sx:{marginTop:"2rem"},children:[(0,a.jsxs)(m.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,a.jsx)(f.Z,{variant:"h4",children:"Set up an evaluation"}),(0,a.jsxs)(m.Z,{direction:"row",spacing:2,children:[(0,a.jsx)(Y,{}),(0,a.jsx)(y,{}),(0,a.jsx)(o.Z,{variant:"outlined",color:"primary",onClick:()=>t(!0),children:"Reset"})]})]}),(0,a.jsx)(s.Z,{mt:4}),(0,a.jsx)(s.Z,{mt:2,children:(0,a.jsx)(r.SV,{FallbackComponent:ec,onReset:()=>{v([])},children:(0,a.jsxs)(m.Z,{direction:"column",spacing:2,justifyContent:"space-between",children:[(0,a.jsx)(f.Z,{variant:"h5",children:"Providers"}),(0,a.jsx)(q,{providers:j,onChange:v})]})})}),(0,a.jsx)(s.Z,{mt:4}),(0,a.jsx)(r.SV,{FallbackComponent:ec,onReset:()=>{b([])},children:(0,a.jsx)(W,{})}),(0,a.jsx)(s.Z,{mt:6}),(0,a.jsx)(r.SV,{FallbackComponent:ec,onReset:()=>{C([])},children:(0,a.jsx)(Q,{varsList:k})}),(0,a.jsx)(es,{}),(0,a.jsxs)(d.Z,{open:e,onClose:()=>t(!1),"aria-labelledby":"alert-dialog-title","aria-describedby":"alert-dialog-description",children:[(0,a.jsx)(x.Z,{id:"alert-dialog-title",children:"Confirm Reset"}),(0,a.jsx)(u.Z,{children:(0,a.jsx)(h.Z,{id:"alert-dialog-description",children:"Are you sure you want to reset all the fields? This action cannot be undone."})}),(0,a.jsxs)(p.Z,{children:[(0,a.jsx)(o.Z,{onClick:()=>t(!1),children:"Cancel"}),(0,a.jsx)(o.Z,{onClick:()=>{g(""),v([]),b([]),C([]),t(!1)},autoFocus:!0,children:"Reset"})]})]})]})}},52428:function(e,t,n){"use strict";n.d(t,{Ox:function(){return r},T8:function(){return i},eA:function(){return l}});var a=n(77580);let i=!a.env.NEXT_PUBLIC_PROMPTFOO_BUILD_STANDALONE_SERVER,r=!!a.env.NEXT_PUBLIC_PROMPTFOO_USE_SUPABASE,l=""},47887:function(e,t,n){"use strict";n.d(t,{o:function(){return r}});var a=n(79685),i=n(74595);let r=(0,a.Ue)()((0,i.tJ)((e,t)=>({env:{},testCases:[],description:"",providers:[],prompts:[],defaultTest:{},evaluateOptions:{},setEnv:t=>e({env:t}),setTestCases:t=>e({testCases:t}),setDescription:t=>e({description:t}),setProviders:t=>e({providers:t}),setPrompts:t=>e({prompts:t}),setDefaultTest:t=>e({defaultTest:t}),setEvaluateOptions:t=>e({evaluateOptions:t}),setStateFromConfig:t=>{let n={};t.description&&(n.description=t.description||""),t.tests&&(n.testCases=t.tests),t.providers&&(n.providers=t.providers),t.prompts&&("string"==typeof t.prompts?n.prompts=[t.prompts]:Array.isArray(t.prompts)?n.prompts=t.prompts.filter(e=>"string"==typeof e&&!e.endsWith(".txt")&&!e.endsWith(".json")&&!e.endsWith(".yaml")):console.warn("Invalid prompts config",t.prompts)),t.defaultTest&&(n.defaultTest=t.defaultTest),t.evaluateOptions&&(n.evaluateOptions=t.evaluateOptions),e(n)},getTestSuite:()=>{let{description:e,testCases:n,providers:a,prompts:i,env:r}=t();return{env:r,description:e,providers:a,prompts:i,tests:n}}}),{name:"promptfoo",skipHydration:!0}))},94499:function(){},55096:function(){},7517:function(){},15320:function(e,t){"use strict";t.It=void 0,t.It=function(e){let t,n,a,i;let r={},l=[],s={};for(let[o,c]of Object.entries(e))o.startsWith("__expected")?""!==c.trim()&&l.push(function(e){if(e.startsWith("javascript:")||e.startsWith("fn:")||e.startsWith("eval:")){let t;e.startsWith("javascript:")&&(t=11),e.startsWith("fn:")&&(t=3),e.startsWith("eval:")&&(t=5);let n=e.slice(t).trim();return{type:"javascript",value:n}}if(e.startsWith("grade:")||e.startsWith("llm-rubric:"))return{type:"llm-rubric",value:e.slice(6)};if(e.startsWith("python:")){let t=e.slice(7).trim();return{type:"python",value:t}}let t=e.match(/^(not-)?(equals|contains-any|contains-all|icontains-any|icontains-all|contains-json|is-json|is-sql|regex|icontains|contains|webhook|rouge-n|similar|starts-with|levenshtein|classifier|model-graded-factuality|factuality|model-graded-closedqa|answer-relevance|context-recall|context-relevance|context-faithfulness|is-valid-openai-function-call|is-valid-openai-tools-call|latency|perplexity|perplexity-score|cost)(?:\((\d+(?:\.\d+)?)\))?(?::([\s\S]*))?$/);if(t){let[e,n,a,i,r]=t,l=n?`not-${a}`:a,s=parseFloat(i);return"contains-any"===a||"contains-all"===a||"icontains-any"===a||"icontains-all"===a?{type:l,value:r.split(",").map(e=>e.trim())}:"contains-json"===a||"is-json"===a?{type:l,value:r}:"rouge-n"===a||"similar"===a||"starts-with"===a||"levenshtein"===a||"classifier"===a||"answer-relevance"===a||"context-recall"===a||"context-relevance"===a||"context-faithfulness"===a||"latency"===a||"perplexity"===a||"perplexity-score"===a||"cost"===a?{type:l,value:r,threshold:s||("similar"===a?.8:.75)}:{type:l,value:r}}return{type:"equals",value:e}}(c)):"__prefix"===o?s.prefix=c:"__suffix"===o?s.suffix=c:"__description"===o?n=c:"__providerOutput"===o?t=c:"__metric"===o?a=c:"__threshold"===o?i=parseFloat(c):r[o]=c;for(let e of l)e.metric=a;return{vars:r,assert:l,options:s,...n?{description:n}:{},...t?{providerOutput:t}:{},...i?{threshold:i}:{}}}}},function(e){e.O(0,[540,640,215,620,770,304,339,995,378,374,235,873,106,258,470,730,744],function(){return e(e.s=76178)}),_N_E=e.O()}]);
@@ -0,0 +1 @@
1
+ !function(){"use strict";var e,t,n,r,o,u,i,c,f,a={},l={};function s(e){var t=l[e];if(void 0!==t)return t.exports;var n=l[e]={exports:{}},r=!0;try{a[e].call(n.exports,n,n.exports,s),r=!1}finally{r&&delete l[e]}return n.exports}s.m=a,e=[],s.O=function(t,n,r,o){if(n){o=o||0;for(var u=e.length;u>0&&e[u-1][2]>o;u--)e[u]=e[u-1];e[u]=[n,r,o];return}for(var i=1/0,u=0;u<e.length;u++){for(var n=e[u][0],r=e[u][1],o=e[u][2],c=!0,f=0;f<n.length;f++)i>=o&&Object.keys(s.O).every(function(e){return s.O[e](n[f])})?n.splice(f--,1):(c=!1,o<i&&(i=o));if(c){e.splice(u--,1);var a=r();void 0!==a&&(t=a)}}return t},s.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return s.d(t,{a:t}),t},n=Object.getPrototypeOf?function(e){return Object.getPrototypeOf(e)}:function(e){return e.__proto__},s.t=function(e,r){if(1&r&&(e=this(e)),8&r||"object"==typeof e&&e&&(4&r&&e.__esModule||16&r&&"function"==typeof e.then))return e;var o=Object.create(null);s.r(o);var u={};t=t||[null,n({}),n([]),n(n)];for(var i=2&r&&e;"object"==typeof i&&!~t.indexOf(i);i=n(i))Object.getOwnPropertyNames(i).forEach(function(t){u[t]=function(){return e[t]}});return u.default=function(){return e},s.d(o,u),o},s.d=function(e,t){for(var n in t)s.o(t,n)&&!s.o(e,n)&&Object.defineProperty(e,n,{enumerable:!0,get:t[n]})},s.f={},s.e=function(e){return Promise.all(Object.keys(s.f).reduce(function(t,n){return s.f[n](e,t),t},[]))},s.u=function(e){return 631===e?"static/chunks/631-87fea556bbafe972.js":"static/chunks/"+e+"."+({407:"42b5daf65ab5dde9",887:"028187fb318c3e59"})[e]+".js"},s.miniCssF=function(e){return"static/css/"+({11:"451beaa5570cb9d3",185:"b76f1739b5a14901",413:"edcd6f0b6c902fde",420:"dab5d695b3657d59",421:"e9f25719d0b14939",466:"48d388184a2f4ce3",931:"8119d8bd13a8adab"})[e]+".css"},s.g=function(){if("object"==typeof globalThis)return globalThis;try{return this||Function("return this")()}catch(e){if("object"==typeof window)return window}}(),s.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},r={},o="_N_E:",s.l=function(e,t,n,u){if(r[e]){r[e].push(t);return}if(void 0!==n)for(var i,c,f=document.getElementsByTagName("script"),a=0;a<f.length;a++){var l=f[a];if(l.getAttribute("src")==e||l.getAttribute("data-webpack")==o+n){i=l;break}}i||(c=!0,(i=document.createElement("script")).charset="utf-8",i.timeout=120,s.nc&&i.setAttribute("nonce",s.nc),i.setAttribute("data-webpack",o+n),i.src=s.tu(e)),r[e]=[t];var d=function(t,n){i.onerror=i.onload=null,clearTimeout(p);var o=r[e];if(delete r[e],i.parentNode&&i.parentNode.removeChild(i),o&&o.forEach(function(e){return e(n)}),t)return t(n)},p=setTimeout(d.bind(null,void 0,{type:"timeout",target:i}),12e4);i.onerror=d.bind(null,i.onerror),i.onload=d.bind(null,i.onload),c&&document.head.appendChild(i)},s.r=function(e){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})},s.tt=function(){return void 0===u&&(u={createScriptURL:function(e){return e}},"undefined"!=typeof trustedTypes&&trustedTypes.createPolicy&&(u=trustedTypes.createPolicy("nextjs#bundler",u))),u},s.tu=function(e){return s.tt().createScriptURL(e)},s.p="/_next/",i={272:0},s.f.j=function(e,t){var n=s.o(i,e)?i[e]:void 0;if(0!==n){if(n)t.push(n[2]);else if(272!=e){var r=new Promise(function(t,r){n=i[e]=[t,r]});t.push(n[2]=r);var o=s.p+s.u(e),u=Error();s.l(o,function(t){if(s.o(i,e)&&(0!==(n=i[e])&&(i[e]=void 0),n)){var r=t&&("load"===t.type?"missing":t.type),o=t&&t.target&&t.target.src;u.message="Loading chunk "+e+" failed.\n("+r+": "+o+")",u.name="ChunkLoadError",u.type=r,u.request=o,n[1](u)}},"chunk-"+e,e)}else i[e]=0}},s.O.j=function(e){return 0===i[e]},c=function(e,t){var n,r,o=t[0],u=t[1],c=t[2],f=0;if(o.some(function(e){return 0!==i[e]})){for(n in u)s.o(u,n)&&(s.m[n]=u[n]);if(c)var a=c(s)}for(e&&e(t);f<o.length;f++)r=o[f],s.o(i,r)&&i[r]&&i[r][0](),i[r]=0;return s.O(a)},(f=self.webpackChunk_N_E=self.webpackChunk_N_E||[]).forEach(c.bind(null,0)),f.push=c.bind(null,f.push.bind(f))}();