node-llama-cpp 3.0.0-beta.1 → 3.0.0-beta.11

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 (354) hide show
  1. package/README.md +2 -0
  2. package/dist/ChatWrapper.d.ts +49 -0
  3. package/dist/ChatWrapper.js +120 -0
  4. package/dist/ChatWrapper.js.map +1 -0
  5. package/dist/{utils/getBin.d.ts → bindings/AddonTypes.d.ts} +14 -4
  6. package/dist/bindings/AddonTypes.js +2 -0
  7. package/dist/bindings/AddonTypes.js.map +1 -0
  8. package/dist/bindings/Llama.d.ts +23 -0
  9. package/dist/bindings/Llama.js +225 -0
  10. package/dist/bindings/Llama.js.map +1 -0
  11. package/dist/bindings/getLlama.d.ts +103 -0
  12. package/dist/bindings/getLlama.js +228 -0
  13. package/dist/bindings/getLlama.js.map +1 -0
  14. package/dist/bindings/types.d.ts +33 -0
  15. package/dist/bindings/types.js +30 -0
  16. package/dist/bindings/types.js.map +1 -0
  17. package/dist/bindings/utils/NoBinaryFoundError.d.ts +2 -0
  18. package/dist/bindings/utils/NoBinaryFoundError.js +7 -0
  19. package/dist/bindings/utils/NoBinaryFoundError.js.map +1 -0
  20. package/dist/{utils → bindings/utils}/binariesGithubRelease.js +1 -1
  21. package/dist/bindings/utils/binariesGithubRelease.js.map +1 -0
  22. package/dist/bindings/utils/clearAllLocalBuilds.d.ts +1 -0
  23. package/dist/bindings/utils/clearAllLocalBuilds.js +47 -0
  24. package/dist/bindings/utils/clearAllLocalBuilds.js.map +1 -0
  25. package/dist/bindings/utils/cloneLlamaCppRepo.d.ts +11 -0
  26. package/dist/bindings/utils/cloneLlamaCppRepo.js +155 -0
  27. package/dist/bindings/utils/cloneLlamaCppRepo.js.map +1 -0
  28. package/dist/bindings/utils/compileLLamaCpp.d.ts +12 -0
  29. package/dist/bindings/utils/compileLLamaCpp.js +157 -0
  30. package/dist/bindings/utils/compileLLamaCpp.js.map +1 -0
  31. package/dist/bindings/utils/getBuildFolderNameForBuildOptions.d.ts +5 -0
  32. package/dist/bindings/utils/getBuildFolderNameForBuildOptions.js +85 -0
  33. package/dist/bindings/utils/getBuildFolderNameForBuildOptions.js.map +1 -0
  34. package/dist/bindings/utils/getCanUsePrebuiltBinaries.d.ts +1 -0
  35. package/dist/bindings/utils/getCanUsePrebuiltBinaries.js +8 -0
  36. package/dist/bindings/utils/getCanUsePrebuiltBinaries.js.map +1 -0
  37. package/dist/bindings/utils/getExampleUsageCodeOfGetLlama.d.ts +2 -0
  38. package/dist/bindings/utils/getExampleUsageCodeOfGetLlama.js +21 -0
  39. package/dist/bindings/utils/getExampleUsageCodeOfGetLlama.js.map +1 -0
  40. package/dist/bindings/utils/getPlatform.d.ts +2 -0
  41. package/dist/bindings/utils/getPlatform.js +15 -0
  42. package/dist/bindings/utils/getPlatform.js.map +1 -0
  43. package/dist/bindings/utils/lastBuildInfo.d.ts +6 -0
  44. package/dist/bindings/utils/lastBuildInfo.js +17 -0
  45. package/dist/bindings/utils/lastBuildInfo.js.map +1 -0
  46. package/dist/bindings/utils/logBinaryUsageExampleToConsole.d.ts +2 -0
  47. package/dist/bindings/utils/logBinaryUsageExampleToConsole.js +28 -0
  48. package/dist/bindings/utils/logBinaryUsageExampleToConsole.js.map +1 -0
  49. package/dist/bindings/utils/resolveChatWrapperBasedOnWrapperTypeName.d.ts +26 -0
  50. package/dist/bindings/utils/resolveChatWrapperBasedOnWrapperTypeName.js +43 -0
  51. package/dist/bindings/utils/resolveChatWrapperBasedOnWrapperTypeName.js.map +1 -0
  52. package/dist/bindings/utils/resolveCustomCmakeOptions.d.ts +1 -0
  53. package/dist/bindings/utils/resolveCustomCmakeOptions.js +43 -0
  54. package/dist/bindings/utils/resolveCustomCmakeOptions.js.map +1 -0
  55. package/dist/chatWrappers/AlpacaChatWrapper.d.ts +12 -0
  56. package/dist/chatWrappers/AlpacaChatWrapper.js +21 -0
  57. package/dist/chatWrappers/AlpacaChatWrapper.js.map +1 -0
  58. package/dist/chatWrappers/ChatMLChatWrapper.d.ts +18 -0
  59. package/dist/chatWrappers/ChatMLChatWrapper.js +83 -0
  60. package/dist/chatWrappers/ChatMLChatWrapper.js.map +1 -0
  61. package/dist/chatWrappers/EmptyChatWrapper.d.ts +4 -0
  62. package/dist/chatWrappers/EmptyChatWrapper.js +5 -0
  63. package/dist/chatWrappers/EmptyChatWrapper.js.map +1 -0
  64. package/dist/chatWrappers/FalconChatWrapper.d.ts +21 -0
  65. package/dist/chatWrappers/FalconChatWrapper.js +104 -0
  66. package/dist/chatWrappers/FalconChatWrapper.js.map +1 -0
  67. package/dist/chatWrappers/FunctionaryChatWrapper.d.ts +41 -0
  68. package/dist/chatWrappers/FunctionaryChatWrapper.js +200 -0
  69. package/dist/chatWrappers/FunctionaryChatWrapper.js.map +1 -0
  70. package/dist/chatWrappers/GeneralChatWrapper.d.ts +21 -0
  71. package/dist/chatWrappers/GeneralChatWrapper.js +112 -0
  72. package/dist/chatWrappers/GeneralChatWrapper.js.map +1 -0
  73. package/dist/chatWrappers/LlamaChatWrapper.d.ts +13 -0
  74. package/dist/chatWrappers/LlamaChatWrapper.js +78 -0
  75. package/dist/chatWrappers/LlamaChatWrapper.js.map +1 -0
  76. package/dist/chatWrappers/resolveChatWrapperBasedOnModel.d.ts +5 -5
  77. package/dist/chatWrappers/resolveChatWrapperBasedOnModel.js +28 -17
  78. package/dist/chatWrappers/resolveChatWrapperBasedOnModel.js.map +1 -1
  79. package/dist/cli/cli.js +8 -0
  80. package/dist/cli/cli.js.map +1 -1
  81. package/dist/cli/commands/BuildCommand.d.ts +2 -1
  82. package/dist/cli/commands/BuildCommand.js +50 -10
  83. package/dist/cli/commands/BuildCommand.js.map +1 -1
  84. package/dist/cli/commands/ChatCommand.d.ts +11 -4
  85. package/dist/cli/commands/ChatCommand.js +138 -64
  86. package/dist/cli/commands/ChatCommand.js.map +1 -1
  87. package/dist/cli/commands/ClearCommand.js +4 -6
  88. package/dist/cli/commands/ClearCommand.js.map +1 -1
  89. package/dist/cli/commands/CompleteCommand.d.ts +25 -0
  90. package/dist/cli/commands/CompleteCommand.js +278 -0
  91. package/dist/cli/commands/CompleteCommand.js.map +1 -0
  92. package/dist/cli/commands/DebugCommand.d.ts +7 -0
  93. package/dist/cli/commands/DebugCommand.js +59 -0
  94. package/dist/cli/commands/DebugCommand.js.map +1 -0
  95. package/dist/cli/commands/DownloadCommand.d.ts +2 -1
  96. package/dist/cli/commands/DownloadCommand.js +47 -40
  97. package/dist/cli/commands/DownloadCommand.js.map +1 -1
  98. package/dist/cli/commands/InfillCommand.d.ts +27 -0
  99. package/dist/cli/commands/InfillCommand.js +316 -0
  100. package/dist/cli/commands/InfillCommand.js.map +1 -0
  101. package/dist/cli/commands/OnPostInstallCommand.js +7 -10
  102. package/dist/cli/commands/OnPostInstallCommand.js.map +1 -1
  103. package/dist/config.d.ts +10 -3
  104. package/dist/config.js +18 -7
  105. package/dist/config.js.map +1 -1
  106. package/dist/consts.d.ts +1 -0
  107. package/dist/consts.js +2 -0
  108. package/dist/consts.js.map +1 -0
  109. package/dist/evaluator/LlamaChat/LlamaChat.d.ts +154 -0
  110. package/dist/evaluator/LlamaChat/LlamaChat.js +684 -0
  111. package/dist/evaluator/LlamaChat/LlamaChat.js.map +1 -0
  112. package/dist/evaluator/LlamaChat/utils/FunctionCallGrammar.d.ts +22 -0
  113. package/dist/evaluator/LlamaChat/utils/FunctionCallGrammar.js +121 -0
  114. package/dist/evaluator/LlamaChat/utils/FunctionCallGrammar.js.map +1 -0
  115. package/dist/evaluator/LlamaChat/utils/contextShiftStrategies/eraseFirstResponseAndKeepFirstSystemChatContextShiftStrategy.d.ts +16 -0
  116. package/dist/evaluator/LlamaChat/utils/contextShiftStrategies/eraseFirstResponseAndKeepFirstSystemChatContextShiftStrategy.js +135 -0
  117. package/dist/evaluator/LlamaChat/utils/contextShiftStrategies/eraseFirstResponseAndKeepFirstSystemChatContextShiftStrategy.js.map +1 -0
  118. package/dist/{llamaEvaluator → evaluator/LlamaChatSession}/LlamaChatSession.d.ts +59 -25
  119. package/dist/evaluator/LlamaChatSession/LlamaChatSession.js +219 -0
  120. package/dist/evaluator/LlamaChatSession/LlamaChatSession.js.map +1 -0
  121. package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.d.ts +7 -0
  122. package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.js +8 -0
  123. package/dist/evaluator/LlamaChatSession/utils/defineChatSessionFunction.js.map +1 -0
  124. package/dist/evaluator/LlamaCompletion.d.ts +148 -0
  125. package/dist/evaluator/LlamaCompletion.js +402 -0
  126. package/dist/evaluator/LlamaCompletion.js.map +1 -0
  127. package/dist/{llamaEvaluator → evaluator}/LlamaContext/LlamaContext.d.ts +20 -23
  128. package/dist/{llamaEvaluator → evaluator}/LlamaContext/LlamaContext.js +77 -107
  129. package/dist/evaluator/LlamaContext/LlamaContext.js.map +1 -0
  130. package/dist/{llamaEvaluator → evaluator}/LlamaContext/types.d.ts +6 -14
  131. package/dist/evaluator/LlamaContext/types.js.map +1 -0
  132. package/dist/evaluator/LlamaContext/utils/batchItemsPrioritizingStrategies/firstInFirstOutStrategy.js.map +1 -0
  133. package/dist/evaluator/LlamaContext/utils/batchItemsPrioritizingStrategies/maximumParallelismStrategy.js.map +1 -0
  134. package/dist/evaluator/LlamaContext/utils/resolveBatchItemsPrioritizingStrategy.js.map +1 -0
  135. package/dist/evaluator/LlamaEmbeddingContext.d.ts +37 -0
  136. package/dist/evaluator/LlamaEmbeddingContext.js +78 -0
  137. package/dist/evaluator/LlamaEmbeddingContext.js.map +1 -0
  138. package/dist/evaluator/LlamaGrammar.d.ts +30 -0
  139. package/dist/{llamaEvaluator → evaluator}/LlamaGrammar.js +14 -18
  140. package/dist/evaluator/LlamaGrammar.js.map +1 -0
  141. package/dist/{llamaEvaluator → evaluator}/LlamaGrammarEvaluationState.js +4 -4
  142. package/dist/evaluator/LlamaGrammarEvaluationState.js.map +1 -0
  143. package/dist/{llamaEvaluator → evaluator}/LlamaJsonSchemaGrammar.d.ts +2 -1
  144. package/dist/{llamaEvaluator → evaluator}/LlamaJsonSchemaGrammar.js +4 -2
  145. package/dist/evaluator/LlamaJsonSchemaGrammar.js.map +1 -0
  146. package/dist/{llamaEvaluator → evaluator}/LlamaModel.d.ts +24 -6
  147. package/dist/{llamaEvaluator → evaluator}/LlamaModel.js +51 -12
  148. package/dist/evaluator/LlamaModel.js.map +1 -0
  149. package/dist/index.d.ts +30 -17
  150. package/dist/index.js +29 -15
  151. package/dist/index.js.map +1 -1
  152. package/dist/state.d.ts +2 -0
  153. package/dist/state.js +7 -0
  154. package/dist/state.js.map +1 -1
  155. package/dist/types.d.ts +72 -3
  156. package/dist/types.js +5 -1
  157. package/dist/types.js.map +1 -1
  158. package/dist/utils/LlamaText.d.ts +42 -0
  159. package/dist/utils/LlamaText.js +207 -0
  160. package/dist/utils/LlamaText.js.map +1 -0
  161. package/dist/utils/StopGenerationDetector.d.ts +28 -0
  162. package/dist/utils/StopGenerationDetector.js +205 -0
  163. package/dist/utils/StopGenerationDetector.js.map +1 -0
  164. package/dist/utils/TokenStreamRegulator.d.ts +30 -0
  165. package/dist/utils/TokenStreamRegulator.js +96 -0
  166. package/dist/utils/TokenStreamRegulator.js.map +1 -0
  167. package/dist/utils/UnsupportedError.d.ts +2 -0
  168. package/dist/utils/UnsupportedError.js +7 -0
  169. package/dist/utils/UnsupportedError.js.map +1 -0
  170. package/dist/utils/appendUserMessageToChatHistory.d.ts +2 -0
  171. package/dist/utils/appendUserMessageToChatHistory.js +18 -0
  172. package/dist/utils/appendUserMessageToChatHistory.js.map +1 -0
  173. package/dist/utils/cmake.js +16 -11
  174. package/dist/utils/cmake.js.map +1 -1
  175. package/dist/utils/compareTokens.d.ts +2 -0
  176. package/dist/utils/compareTokens.js +4 -0
  177. package/dist/utils/compareTokens.js.map +1 -0
  178. package/dist/utils/findCharacterRemovalCountToFitChatHistoryInContext.d.ts +18 -0
  179. package/dist/utils/findCharacterRemovalCountToFitChatHistoryInContext.js +61 -0
  180. package/dist/utils/findCharacterRemovalCountToFitChatHistoryInContext.js.map +1 -0
  181. package/dist/utils/gbnfJson/GbnfGrammarGenerator.d.ts +1 -0
  182. package/dist/utils/gbnfJson/GbnfGrammarGenerator.js +17 -0
  183. package/dist/utils/gbnfJson/GbnfGrammarGenerator.js.map +1 -1
  184. package/dist/utils/gbnfJson/GbnfTerminal.d.ts +1 -1
  185. package/dist/utils/gbnfJson/GbnfTerminal.js.map +1 -1
  186. package/dist/utils/gbnfJson/terminals/GbnfVerbatimText.d.ts +6 -0
  187. package/dist/utils/gbnfJson/terminals/GbnfVerbatimText.js +21 -0
  188. package/dist/utils/gbnfJson/terminals/GbnfVerbatimText.js.map +1 -0
  189. package/dist/utils/gbnfJson/types.d.ts +1 -1
  190. package/dist/utils/gbnfJson/types.js.map +1 -1
  191. package/dist/utils/gbnfJson/utils/validateObjectAgainstGbnfSchema.d.ts +1 -0
  192. package/dist/utils/gbnfJson/utils/validateObjectAgainstGbnfSchema.js.map +1 -1
  193. package/dist/utils/getConsoleLogPrefix.d.ts +1 -0
  194. package/dist/utils/getConsoleLogPrefix.js +9 -0
  195. package/dist/utils/getConsoleLogPrefix.js.map +1 -0
  196. package/dist/utils/getGbnfGrammarForGbnfJsonSchema.js +1 -15
  197. package/dist/utils/getGbnfGrammarForGbnfJsonSchema.js.map +1 -1
  198. package/dist/utils/getGrammarsFolder.d.ts +2 -1
  199. package/dist/utils/getGrammarsFolder.js +8 -7
  200. package/dist/utils/getGrammarsFolder.js.map +1 -1
  201. package/dist/utils/getModuleVersion.d.ts +1 -0
  202. package/dist/utils/getModuleVersion.js +13 -0
  203. package/dist/utils/getModuleVersion.js.map +1 -0
  204. package/dist/utils/getQueuedTokensBeforeStopTrigger.d.ts +6 -0
  205. package/dist/utils/getQueuedTokensBeforeStopTrigger.js +22 -0
  206. package/dist/utils/getQueuedTokensBeforeStopTrigger.js.map +1 -0
  207. package/dist/utils/getTypeScriptTypeStringForGbnfJsonSchema.d.ts +2 -0
  208. package/dist/utils/getTypeScriptTypeStringForGbnfJsonSchema.js +49 -0
  209. package/dist/utils/getTypeScriptTypeStringForGbnfJsonSchema.js.map +1 -0
  210. package/dist/utils/gitReleaseBundles.js +6 -5
  211. package/dist/utils/gitReleaseBundles.js.map +1 -1
  212. package/dist/utils/hashString.d.ts +1 -0
  213. package/dist/utils/hashString.js +8 -0
  214. package/dist/utils/hashString.js.map +1 -0
  215. package/dist/utils/isLockfileActive.d.ts +4 -0
  216. package/dist/utils/isLockfileActive.js +12 -0
  217. package/dist/utils/isLockfileActive.js.map +1 -0
  218. package/dist/utils/parseModelTypeDescription.d.ts +1 -1
  219. package/dist/utils/prettyPrintObject.d.ts +1 -0
  220. package/dist/utils/prettyPrintObject.js +40 -0
  221. package/dist/utils/prettyPrintObject.js.map +1 -0
  222. package/dist/utils/removeNullFields.d.ts +1 -0
  223. package/dist/utils/removeNullFields.js +8 -0
  224. package/dist/utils/removeNullFields.js.map +1 -1
  225. package/dist/utils/resolveChatWrapper.d.ts +4 -0
  226. package/dist/utils/resolveChatWrapper.js +16 -0
  227. package/dist/utils/resolveChatWrapper.js.map +1 -0
  228. package/dist/utils/resolveGithubRelease.d.ts +2 -0
  229. package/dist/utils/resolveGithubRelease.js +36 -0
  230. package/dist/utils/resolveGithubRelease.js.map +1 -0
  231. package/dist/utils/spawnCommand.d.ts +1 -1
  232. package/dist/utils/spawnCommand.js +4 -2
  233. package/dist/utils/spawnCommand.js.map +1 -1
  234. package/dist/utils/tokenizeInput.d.ts +3 -0
  235. package/dist/utils/tokenizeInput.js +9 -0
  236. package/dist/utils/tokenizeInput.js.map +1 -0
  237. package/dist/utils/truncateTextAndRoundToWords.d.ts +8 -0
  238. package/dist/utils/truncateTextAndRoundToWords.js +27 -0
  239. package/dist/utils/truncateTextAndRoundToWords.js.map +1 -0
  240. package/dist/utils/waitForLockfileRelease.d.ts +5 -0
  241. package/dist/utils/waitForLockfileRelease.js +20 -0
  242. package/dist/utils/waitForLockfileRelease.js.map +1 -0
  243. package/dist/utils/withLockfile.d.ts +7 -0
  244. package/dist/utils/withLockfile.js +44 -0
  245. package/dist/utils/withLockfile.js.map +1 -0
  246. package/dist/utils/withOra.js +11 -1
  247. package/dist/utils/withOra.js.map +1 -1
  248. package/dist/utils/withStatusLogs.d.ts +2 -1
  249. package/dist/utils/withStatusLogs.js +11 -8
  250. package/dist/utils/withStatusLogs.js.map +1 -1
  251. package/llama/.clang-format +1 -2
  252. package/llama/CMakeLists.txt +87 -2
  253. package/llama/addon.cpp +319 -31
  254. package/llama/binariesGithubRelease.json +1 -1
  255. package/llama/gitRelease.bundle +0 -0
  256. package/llama/gpuInfo/cuda-gpu-info.cu +99 -0
  257. package/llama/gpuInfo/cuda-gpu-info.h +7 -0
  258. package/llama/gpuInfo/metal-gpu-info.h +5 -0
  259. package/llama/gpuInfo/metal-gpu-info.mm +17 -0
  260. package/llama/llama.cpp.info.json +4 -0
  261. package/llamaBins/linux-arm64/.buildMetadata.json +1 -0
  262. package/llamaBins/linux-arm64/llama-addon.node +0 -0
  263. package/llamaBins/linux-armv7l/.buildMetadata.json +1 -0
  264. package/llamaBins/linux-armv7l/llama-addon.node +0 -0
  265. package/llamaBins/linux-x64/.buildMetadata.json +1 -0
  266. package/llamaBins/linux-x64/llama-addon.node +0 -0
  267. package/llamaBins/linux-x64-cuda/.buildMetadata.json +1 -0
  268. package/llamaBins/linux-x64-cuda/llama-addon.node +0 -0
  269. package/llamaBins/mac-arm64-metal/.buildMetadata.json +1 -0
  270. package/llamaBins/mac-arm64-metal/ggml-metal.metal +6491 -0
  271. package/llamaBins/mac-arm64-metal/llama-addon.node +0 -0
  272. package/llamaBins/mac-x64/.buildMetadata.json +1 -0
  273. package/llamaBins/mac-x64/llama-addon.node +0 -0
  274. package/llamaBins/win-x64/.buildMetadata.json +1 -0
  275. package/llamaBins/win-x64/llama-addon.exp +0 -0
  276. package/llamaBins/win-x64/llama-addon.node +0 -0
  277. package/llamaBins/win-x64-cuda/.buildMetadata.json +1 -0
  278. package/llamaBins/win-x64-cuda/llama-addon.exp +0 -0
  279. package/llamaBins/win-x64-cuda/llama-addon.lib +0 -0
  280. package/llamaBins/win-x64-cuda/llama-addon.node +0 -0
  281. package/package.json +37 -15
  282. package/dist/AbortError.d.ts +0 -2
  283. package/dist/AbortError.js +0 -7
  284. package/dist/AbortError.js.map +0 -1
  285. package/dist/ChatPromptWrapper.d.ts +0 -11
  286. package/dist/ChatPromptWrapper.js +0 -20
  287. package/dist/ChatPromptWrapper.js.map +0 -1
  288. package/dist/chatWrappers/ChatMLChatPromptWrapper.d.ts +0 -12
  289. package/dist/chatWrappers/ChatMLChatPromptWrapper.js +0 -22
  290. package/dist/chatWrappers/ChatMLChatPromptWrapper.js.map +0 -1
  291. package/dist/chatWrappers/EmptyChatPromptWrapper.d.ts +0 -4
  292. package/dist/chatWrappers/EmptyChatPromptWrapper.js +0 -5
  293. package/dist/chatWrappers/EmptyChatPromptWrapper.js.map +0 -1
  294. package/dist/chatWrappers/FalconChatPromptWrapper.d.ts +0 -19
  295. package/dist/chatWrappers/FalconChatPromptWrapper.js +0 -33
  296. package/dist/chatWrappers/FalconChatPromptWrapper.js.map +0 -1
  297. package/dist/chatWrappers/GeneralChatPromptWrapper.d.ts +0 -19
  298. package/dist/chatWrappers/GeneralChatPromptWrapper.js +0 -38
  299. package/dist/chatWrappers/GeneralChatPromptWrapper.js.map +0 -1
  300. package/dist/chatWrappers/LlamaChatPromptWrapper.d.ts +0 -12
  301. package/dist/chatWrappers/LlamaChatPromptWrapper.js +0 -23
  302. package/dist/chatWrappers/LlamaChatPromptWrapper.js.map +0 -1
  303. package/dist/chatWrappers/generateContextTextFromConversationHistory.d.ts +0 -15
  304. package/dist/chatWrappers/generateContextTextFromConversationHistory.js +0 -39
  305. package/dist/chatWrappers/generateContextTextFromConversationHistory.js.map +0 -1
  306. package/dist/llamaEvaluator/LlamaBins.d.ts +0 -19
  307. package/dist/llamaEvaluator/LlamaBins.js +0 -5
  308. package/dist/llamaEvaluator/LlamaBins.js.map +0 -1
  309. package/dist/llamaEvaluator/LlamaChatSession.js +0 -290
  310. package/dist/llamaEvaluator/LlamaChatSession.js.map +0 -1
  311. package/dist/llamaEvaluator/LlamaContext/LlamaContext.js.map +0 -1
  312. package/dist/llamaEvaluator/LlamaContext/types.js.map +0 -1
  313. package/dist/llamaEvaluator/LlamaContext/utils/batchItemsPrioritizingStrategies/firstInFirstOutStrategy.js.map +0 -1
  314. package/dist/llamaEvaluator/LlamaContext/utils/batchItemsPrioritizingStrategies/maximumParallelismStrategy.js.map +0 -1
  315. package/dist/llamaEvaluator/LlamaContext/utils/resolveBatchItemsPrioritizingStrategy.js.map +0 -1
  316. package/dist/llamaEvaluator/LlamaGrammar.d.ts +0 -32
  317. package/dist/llamaEvaluator/LlamaGrammar.js.map +0 -1
  318. package/dist/llamaEvaluator/LlamaGrammarEvaluationState.js.map +0 -1
  319. package/dist/llamaEvaluator/LlamaJsonSchemaGrammar.js.map +0 -1
  320. package/dist/llamaEvaluator/LlamaModel.js.map +0 -1
  321. package/dist/utils/binariesGithubRelease.js.map +0 -1
  322. package/dist/utils/clearLlamaBuild.d.ts +0 -1
  323. package/dist/utils/clearLlamaBuild.js +0 -12
  324. package/dist/utils/clearLlamaBuild.js.map +0 -1
  325. package/dist/utils/cloneLlamaCppRepo.d.ts +0 -2
  326. package/dist/utils/cloneLlamaCppRepo.js +0 -102
  327. package/dist/utils/cloneLlamaCppRepo.js.map +0 -1
  328. package/dist/utils/compileLLamaCpp.d.ts +0 -8
  329. package/dist/utils/compileLLamaCpp.js +0 -127
  330. package/dist/utils/compileLLamaCpp.js.map +0 -1
  331. package/dist/utils/getBin.js +0 -78
  332. package/dist/utils/getBin.js.map +0 -1
  333. package/dist/utils/getReleaseInfo.d.ts +0 -7
  334. package/dist/utils/getReleaseInfo.js +0 -30
  335. package/dist/utils/getReleaseInfo.js.map +0 -1
  336. package/dist/utils/getTextCompletion.d.ts +0 -3
  337. package/dist/utils/getTextCompletion.js +0 -12
  338. package/dist/utils/getTextCompletion.js.map +0 -1
  339. package/dist/utils/usedBinFlag.d.ts +0 -6
  340. package/dist/utils/usedBinFlag.js +0 -15
  341. package/dist/utils/usedBinFlag.js.map +0 -1
  342. package/llama/usedBin.json +0 -3
  343. package/llamaBins/mac-arm64/ggml-metal.metal +0 -2929
  344. package/llamaBins/mac-arm64/llama-addon.node +0 -0
  345. package/llamaBins/mac-x64/ggml-metal.metal +0 -2929
  346. /package/dist/{utils → bindings/utils}/binariesGithubRelease.d.ts +0 -0
  347. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/types.js +0 -0
  348. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/batchItemsPrioritizingStrategies/firstInFirstOutStrategy.d.ts +0 -0
  349. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/batchItemsPrioritizingStrategies/firstInFirstOutStrategy.js +0 -0
  350. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/batchItemsPrioritizingStrategies/maximumParallelismStrategy.d.ts +0 -0
  351. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/batchItemsPrioritizingStrategies/maximumParallelismStrategy.js +0 -0
  352. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/resolveBatchItemsPrioritizingStrategy.d.ts +0 -0
  353. /package/dist/{llamaEvaluator → evaluator}/LlamaContext/utils/resolveBatchItemsPrioritizingStrategy.js +0 -0
  354. /package/dist/{llamaEvaluator → evaluator}/LlamaGrammarEvaluationState.d.ts +0 -0
@@ -0,0 +1,316 @@
1
+ import * as readline from "readline";
2
+ import process from "process";
3
+ import path from "path";
4
+ import chalk from "chalk";
5
+ import fs from "fs-extra";
6
+ import withStatusLogs from "../../utils/withStatusLogs.js";
7
+ import { getLlama } from "../../bindings/getLlama.js";
8
+ import { LlamaModel } from "../../evaluator/LlamaModel.js";
9
+ import { LlamaContext } from "../../evaluator/LlamaContext/LlamaContext.js";
10
+ import { LlamaLogLevel } from "../../bindings/types.js";
11
+ import { LlamaCompletion } from "../../evaluator/LlamaCompletion.js";
12
+ export const InfillCommand = {
13
+ command: "infill",
14
+ describe: "Generate an infill completion for a given suffix and prefix texts",
15
+ builder(yargs) {
16
+ return yargs
17
+ .option("model", {
18
+ alias: "m",
19
+ type: "string",
20
+ demandOption: true,
21
+ description: "Llama model file to use for the chat",
22
+ group: "Required:"
23
+ })
24
+ .option("systemInfo", {
25
+ alias: "i",
26
+ type: "boolean",
27
+ default: false,
28
+ description: "Print llama.cpp system info",
29
+ group: "Optional:"
30
+ })
31
+ .option("prefix", {
32
+ type: "string",
33
+ description: "First prefix text to automatically load",
34
+ group: "Optional:"
35
+ })
36
+ .option("prefixFile", {
37
+ type: "string",
38
+ description: "Path to a file to load prefix text from automatically",
39
+ group: "Optional:"
40
+ })
41
+ .option("suffix", {
42
+ type: "string",
43
+ description: "First suffix text to automatically load. Requires `prefix` or `prefixFile` to be set",
44
+ group: "Optional:"
45
+ })
46
+ .option("suffixFile", {
47
+ type: "string",
48
+ description: "Path to a file to load suffix text from automatically. Requires `prefix` or `prefixFile` to be set",
49
+ group: "Optional:"
50
+ })
51
+ .option("contextSize", {
52
+ alias: "c",
53
+ type: "number",
54
+ default: 1024 * 4,
55
+ description: "Context size to use for the model context",
56
+ group: "Optional:"
57
+ })
58
+ .option("batchSize", {
59
+ alias: "b",
60
+ type: "number",
61
+ description: "Batch size to use for the model context. The default value is the context size",
62
+ group: "Optional:"
63
+ })
64
+ .option("threads", {
65
+ type: "number",
66
+ default: 6,
67
+ description: "Number of threads to use for the evaluation of tokens",
68
+ group: "Optional:"
69
+ })
70
+ .option("temperature", {
71
+ alias: "t",
72
+ type: "number",
73
+ default: 0,
74
+ description: "Temperature is a hyperparameter that controls the randomness of the generated text. It affects the probability distribution of the model's output tokens. A higher temperature (e.g., 1.5) makes the output more random and creative, while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative. The suggested temperature is 0.8, which provides a balance between randomness and determinism. At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run. Set to `0` to disable.",
75
+ group: "Optional:"
76
+ })
77
+ .option("minP", {
78
+ alias: "mp",
79
+ type: "number",
80
+ default: 0,
81
+ description: "From the next token candidates, discard the percentage of tokens with the lowest probability. For example, if set to `0.05`, 5% of the lowest probability tokens will be discarded. This is useful for generating more high-quality results when using a high temperature. Set to a value between `0` and `1` to enable. Only relevant when `temperature` is set to a value greater than `0`.",
82
+ group: "Optional:"
83
+ })
84
+ .option("topK", {
85
+ alias: "k",
86
+ type: "number",
87
+ default: 40,
88
+ description: "Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation. An integer number between `1` and the size of the vocabulary. Set to `0` to disable (which uses the full vocabulary). Only relevant when `temperature` is set to a value greater than 0.",
89
+ group: "Optional:"
90
+ })
91
+ .option("topP", {
92
+ alias: "p",
93
+ type: "number",
94
+ default: 0.95,
95
+ description: "Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P, and samples the next token only from this set. A float number between `0` and `1`. Set to `1` to disable. Only relevant when `temperature` is set to a value greater than `0`.",
96
+ group: "Optional:"
97
+ })
98
+ .option("gpuLayers", {
99
+ alias: "gl",
100
+ type: "number",
101
+ description: "number of layers to store in VRAM",
102
+ group: "Optional:"
103
+ })
104
+ .option("repeatPenalty", {
105
+ alias: "rp",
106
+ type: "number",
107
+ default: 1.1,
108
+ description: "Prevent the model from repeating the same token too much. Set to `1` to disable.",
109
+ group: "Optional:"
110
+ })
111
+ .option("lastTokensRepeatPenalty", {
112
+ alias: "rpn",
113
+ type: "number",
114
+ default: 64,
115
+ description: "Number of recent tokens generated by the model to apply penalties to repetition of",
116
+ group: "Optional:"
117
+ })
118
+ .option("penalizeRepeatingNewLine", {
119
+ alias: "rpnl",
120
+ type: "boolean",
121
+ default: true,
122
+ description: "Penalize new line tokens. set \"--no-penalizeRepeatingNewLine\" or \"--no-rpnl\" to disable",
123
+ group: "Optional:"
124
+ })
125
+ .option("repeatFrequencyPenalty", {
126
+ alias: "rfp",
127
+ type: "number",
128
+ description: "For n time a token is in the `punishTokens` array, lower its probability by `n * repeatFrequencyPenalty`. Set to a value between `0` and `1` to enable.",
129
+ group: "Optional:"
130
+ })
131
+ .option("repeatPresencePenalty", {
132
+ alias: "rpp",
133
+ type: "number",
134
+ description: "Lower the probability of all the tokens in the `punishTokens` array by `repeatPresencePenalty`. Set to a value between `0` and `1` to enable.",
135
+ group: "Optional:"
136
+ })
137
+ .option("maxTokens", {
138
+ alias: "mt",
139
+ type: "number",
140
+ default: 0,
141
+ description: "Maximum number of tokens to generate in responses. Set to `0` to disable. Set to `-1` to set to the context size",
142
+ group: "Optional:"
143
+ })
144
+ .option("noInfoLog", {
145
+ alias: "nl",
146
+ type: "boolean",
147
+ default: false,
148
+ description: "Disable llama.cpp info logs",
149
+ group: "Optional:"
150
+ })
151
+ .option("printTimings", {
152
+ alias: "pt",
153
+ type: "boolean",
154
+ default: false,
155
+ description: "Print llama.cpp timings after each response",
156
+ group: "Optional:"
157
+ });
158
+ },
159
+ async handler({ model, systemInfo, prefix, prefixFile, suffix, suffixFile, contextSize, batchSize, threads, temperature, minP, topK, topP, gpuLayers, repeatPenalty, lastTokensRepeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noInfoLog, printTimings }) {
160
+ try {
161
+ await RunInfill({
162
+ model, systemInfo, prefix, prefixFile, suffix, suffixFile, contextSize, batchSize,
163
+ threads, temperature, minP, topK, topP, gpuLayers, lastTokensRepeatPenalty,
164
+ repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens,
165
+ noInfoLog, printTimings
166
+ });
167
+ }
168
+ catch (err) {
169
+ console.error(err);
170
+ process.exit(1);
171
+ }
172
+ }
173
+ };
174
+ async function RunInfill({ model: modelArg, systemInfo, prefix, prefixFile, suffix, suffixFile, contextSize, batchSize, threads, temperature, minP, topK, topP, gpuLayers, lastTokensRepeatPenalty, repeatPenalty, penalizeRepeatingNewLine, repeatFrequencyPenalty, repeatPresencePenalty, maxTokens, noInfoLog, printTimings }) {
175
+ if (noInfoLog)
176
+ console.info(`${chalk.yellow("Log level:")} warn`);
177
+ const llama = await getLlama("lastBuild", {
178
+ logLevel: noInfoLog
179
+ ? LlamaLogLevel.warn
180
+ : LlamaLogLevel.debug
181
+ });
182
+ const logBatchSize = batchSize != null;
183
+ if (systemInfo)
184
+ console.log(llama.systemInfo);
185
+ if (prefixFile != null && prefixFile !== "") {
186
+ if (prefix != null && prefix !== "")
187
+ console.warn(chalk.yellow("Both `prefix` and `prefixFile` were specified. `prefixFile` will be used."));
188
+ prefix = await fs.readFile(path.resolve(process.cwd(), prefixFile), "utf8");
189
+ }
190
+ if (suffixFile != null && suffixFile !== "") {
191
+ if (suffix != null && suffix !== "")
192
+ console.warn(chalk.yellow("Both `suffix` and `suffixFile` were specified. `suffixFile` will be used."));
193
+ suffix = await fs.readFile(path.resolve(process.cwd(), suffixFile), "utf8");
194
+ }
195
+ if (suffix != null && prefix == null) {
196
+ console.warn(chalk.yellow("Suffix was specified but no prefix was specified. Suffix will be ignored."));
197
+ suffix = undefined;
198
+ }
199
+ if (batchSize == null)
200
+ batchSize = contextSize;
201
+ else if (batchSize > contextSize) {
202
+ console.warn(chalk.yellow("Batch size is greater than the context size. Batch size will be set to the context size."));
203
+ batchSize = contextSize;
204
+ }
205
+ let initialPrefix = prefix ?? null;
206
+ let initialSuffix = suffix ?? null;
207
+ const model = await withStatusLogs({
208
+ loading: chalk.blue("Loading model"),
209
+ success: chalk.blue("Model loaded"),
210
+ fail: chalk.blue("Failed to load model")
211
+ }, async () => new LlamaModel({
212
+ llama,
213
+ modelPath: path.resolve(process.cwd(), modelArg),
214
+ gpuLayers: gpuLayers != null ? gpuLayers : undefined
215
+ }));
216
+ const context = await withStatusLogs({
217
+ loading: chalk.blue("Creating context"),
218
+ success: chalk.blue("Context created"),
219
+ fail: chalk.blue("Failed to create context")
220
+ }, async () => new LlamaContext({
221
+ model,
222
+ contextSize,
223
+ batchSize,
224
+ threads
225
+ }));
226
+ const completion = new LlamaCompletion({
227
+ contextSequence: context.getSequence()
228
+ });
229
+ await new Promise((accept) => setTimeout(accept, 0)); // wait for logs to finish printing
230
+ console.info(`${chalk.yellow("Context size:")} ${context.contextSize}`);
231
+ if (logBatchSize)
232
+ console.info(`${chalk.yellow("Batch size:")} ${context.batchSize}`);
233
+ console.info(`${chalk.yellow("Train context size:")} ${model.trainContextSize}`);
234
+ console.info(`${chalk.yellow("Model type:")} ${model.typeDescription}`);
235
+ console.info(`${chalk.yellow("Repeat penalty:")} ${repeatPenalty} (apply to last ${lastTokensRepeatPenalty} tokens)`);
236
+ if (repeatFrequencyPenalty != null)
237
+ console.info(`${chalk.yellow("Repeat frequency penalty:")} ${repeatFrequencyPenalty}`);
238
+ if (repeatPresencePenalty != null)
239
+ console.info(`${chalk.yellow("Repeat presence penalty:")} ${repeatPresencePenalty}`);
240
+ if (!penalizeRepeatingNewLine)
241
+ console.info(`${chalk.yellow("Penalize repeating new line:")} disabled`);
242
+ // this is for ora to not interfere with readline
243
+ await new Promise(resolve => setTimeout(resolve, 1));
244
+ if (!completion.infillSupported) {
245
+ console.log(chalk.red("Infill is not supported for this model"));
246
+ process.exit(1);
247
+ }
248
+ const replPrefixHistory = [];
249
+ const replSuffixHistory = [];
250
+ async function getInput(name) {
251
+ const rl = readline.createInterface({
252
+ input: process.stdin,
253
+ output: process.stdout,
254
+ history: name === "Prefix"
255
+ ? replPrefixHistory.slice()
256
+ : replSuffixHistory.slice()
257
+ });
258
+ const res = await new Promise((accept) => rl.question(chalk.yellow(name + "> "), accept));
259
+ rl.close();
260
+ return res;
261
+ }
262
+ // eslint-disable-next-line no-constant-condition
263
+ while (true) {
264
+ const prefixInput = initialPrefix != null
265
+ ? initialPrefix
266
+ : await getInput("Prefix");
267
+ if (initialPrefix != null) {
268
+ console.log(chalk.green("Prefix> ") + initialPrefix);
269
+ initialPrefix = null;
270
+ }
271
+ else
272
+ await replPrefixHistory.push(prefixInput);
273
+ if (prefixInput === ".exit")
274
+ break;
275
+ const suffixInput = initialSuffix != null
276
+ ? initialSuffix
277
+ : await getInput("Suffix");
278
+ if (initialSuffix != null) {
279
+ console.log(chalk.green("Suffix> ") + initialSuffix);
280
+ initialSuffix = null;
281
+ }
282
+ else
283
+ await replSuffixHistory.push(suffixInput);
284
+ if (suffixInput === ".exit")
285
+ break;
286
+ process.stdout.write(chalk.yellow("Infill: "));
287
+ const [startColor, endColor] = chalk.blue("MIDDLE").split("MIDDLE");
288
+ process.stdout.write(startColor);
289
+ await completion.generateInfillCompletion(prefixInput, suffixInput, {
290
+ temperature,
291
+ minP,
292
+ topK,
293
+ topP,
294
+ repeatPenalty: {
295
+ penalty: repeatPenalty,
296
+ frequencyPenalty: repeatFrequencyPenalty != null ? repeatFrequencyPenalty : undefined,
297
+ presencePenalty: repeatPresencePenalty != null ? repeatPresencePenalty : undefined,
298
+ penalizeNewLine: penalizeRepeatingNewLine,
299
+ lastTokens: lastTokensRepeatPenalty
300
+ },
301
+ maxTokens: maxTokens === -1
302
+ ? context.contextSize
303
+ : maxTokens <= 0
304
+ ? undefined
305
+ : maxTokens,
306
+ onToken(chunk) {
307
+ process.stdout.write(model.detokenize(chunk));
308
+ }
309
+ });
310
+ process.stdout.write(endColor);
311
+ console.log();
312
+ if (printTimings)
313
+ await context.printTimings();
314
+ }
315
+ }
316
+ //# sourceMappingURL=InfillCommand.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"InfillCommand.js","sourceRoot":"","sources":["../../../src/cli/commands/InfillCommand.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,QAAQ,MAAM,UAAU,CAAC;AACrC,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,IAAI,MAAM,MAAM,CAAC;AAExB,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAE,MAAM,UAAU,CAAC;AAC1B,OAAO,cAAc,MAAM,+BAA+B,CAAC;AAC3D,OAAO,EAAC,QAAQ,EAAC,MAAM,4BAA4B,CAAC;AACpD,OAAO,EAAC,UAAU,EAAC,MAAM,+BAA+B,CAAC;AACzD,OAAO,EAAC,YAAY,EAAC,MAAM,8CAA8C,CAAC;AAC1E,OAAO,EAAC,aAAa,EAAC,MAAM,yBAAyB,CAAC;AACtD,OAAO,EAAC,eAAe,EAAC,MAAM,oCAAoC,CAAC;AA2BnE,MAAM,CAAC,MAAM,aAAa,GAAyC;IAC/D,OAAO,EAAE,QAAQ;IACjB,QAAQ,EAAE,mEAAmE;IAC7E,OAAO,CAAC,KAAK;QACT,OAAO,KAAK;aACP,MAAM,CAAC,OAAO,EAAE;YACb,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,YAAY,EAAE,IAAI;YAClB,WAAW,EAAE,sCAAsC;YACnD,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,YAAY,EAAE;YAClB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,6BAA6B;YAC1C,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,QAAQ,EAAE;YACd,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,yCAAyC;YACtD,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,YAAY,EAAE;YAClB,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,uDAAuD;YACpE,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,QAAQ,EAAE;YACd,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,sFAAsF;YACnG,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,YAAY,EAAE;YAClB,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,oGAAoG;YACjH,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,aAAa,EAAE;YACnB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,IAAI,GAAG,CAAC;YACjB,WAAW,EAAE,2CAA2C;YACxD,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,WAAW,EAAE;YACjB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,gFAAgF;YAC7F,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,SAAS,EAAE;YACf,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,CAAC;YACV,WAAW,EAAE,uDAAuD;YACpE,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,aAAa,EAAE;YACnB,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,CAAC;YACV,WAAW,EAAE,6jBAA6jB;YAC1kB,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,MAAM,EAAE;YACZ,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,CAAC;YACV,WAAW,EAAE,+XAA+X;YAC5Y,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,MAAM,EAAE;YACZ,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,EAAE;YACX,WAAW,EAAE,4SAA4S;YACzT,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,MAAM,EAAE;YACZ,KAAK,EAAE,GAAG;YACV,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,IAAI;YACb,WAAW,EAAE,qRAAqR;YAClS,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,WAAW,EAAE;YACjB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,mCAAmC;YAChD,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,eAAe,EAAE;YACrB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,GAAG;YACZ,WAAW,EAAE,kFAAkF;YAC/F,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,yBAAyB,EAAE;YAC/B,KAAK,EAAE,KAAK;YACZ,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,EAAE;YACX,WAAW,EAAE,oFAAoF;YACjG,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,0BAA0B,EAAE;YAChC,KAAK,EAAE,MAAM;YACb,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,IAAI;YACb,WAAW,EAAE,6FAA6F;YAC1G,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,wBAAwB,EAAE;YAC9B,KAAK,EAAE,KAAK;YACZ,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,yJAAyJ;YACtK,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,uBAAuB,EAAE;YAC7B,KAAK,EAAE,KAAK;YACZ,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,+IAA+I;YAC5J,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,WAAW,EAAE;YACjB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,CAAC;YACV,WAAW,EAAE,kHAAkH;YAC/H,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,WAAW,EAAE;YACjB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,6BAA6B;YAC1C,KAAK,EAAE,WAAW;SACrB,CAAC;aACD,MAAM,CAAC,cAAc,EAAE;YACpB,KAAK,EAAE,IAAI;YACX,IAAI,EAAE,SAAS;YACf,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,6CAA6C;YAC1D,KAAK,EAAE,WAAW;SACrB,CAAC,CAAC;IACX,CAAC;IACD,KAAK,CAAC,OAAO,CAAC,EACV,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,EAAE,SAAS,EACjF,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAChC,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,uBAAuB,EAAE,wBAAwB,EACjF,sBAAsB,EAAE,qBAAqB,EAAE,SAAS,EACxD,SAAS,EAAE,YAAY,EAC1B;QACG,IAAI;YACA,MAAM,SAAS,CAAC;gBACZ,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,EAAE,SAAS;gBACjF,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,uBAAuB;gBAC1E,aAAa,EAAE,wBAAwB,EAAE,sBAAsB,EAAE,qBAAqB,EAAE,SAAS;gBACjG,SAAS,EAAE,YAAY;aAC1B,CAAC,CAAC;SACN;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACnB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;SACnB;IACL,CAAC;CACJ,CAAC;AAGF,KAAK,UAAU,SAAS,CAAC,EACrB,KAAK,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,WAAW,EAAE,SAAS,EAC3F,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EACjD,uBAAuB,EAAE,aAAa,EAAE,wBAAwB,EAAE,sBAAsB,EAAE,qBAAqB,EAC/G,SAAS,EAAE,SAAS,EAAE,YAAY,EACtB;IACZ,IAAI,SAAS;QACT,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC;IAEvD,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC,WAAW,EAAE;QACtC,QAAQ,EAAE,SAAS;YACf,CAAC,CAAC,aAAa,CAAC,IAAI;YACpB,CAAC,CAAC,aAAa,CAAC,KAAK;KAC5B,CAAC,CAAC;IACH,MAAM,YAAY,GAAG,SAAS,IAAI,IAAI,CAAC;IAEvC,IAAI,UAAU;QACV,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IAElC,IAAI,UAAU,IAAI,IAAI,IAAI,UAAU,KAAK,EAAE,EAAE;QACzC,IAAI,MAAM,IAAI,IAAI,IAAI,MAAM,KAAK,EAAE;YAC/B,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,2EAA2E,CAAC,CAAC,CAAC;QAE5G,MAAM,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC;KAC/E;IAED,IAAI,UAAU,IAAI,IAAI,IAAI,UAAU,KAAK,EAAE,EAAE;QACzC,IAAI,MAAM,IAAI,IAAI,IAAI,MAAM,KAAK,EAAE;YAC/B,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,2EAA2E,CAAC,CAAC,CAAC;QAE5G,MAAM,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC;KAC/E;IAED,IAAI,MAAM,IAAI,IAAI,IAAI,MAAM,IAAI,IAAI,EAAE;QAClC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,2EAA2E,CAAC,CAAC,CAAC;QACxG,MAAM,GAAG,SAAS,CAAC;KACtB;IAED,IAAI,SAAS,IAAI,IAAI;QACjB,SAAS,GAAG,WAAW,CAAC;SACvB,IAAI,SAAS,GAAG,WAAW,EAAE;QAC9B,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,0FAA0F,CAAC,CAAC,CAAC;QACvH,SAAS,GAAG,WAAW,CAAC;KAC3B;IAED,IAAI,aAAa,GAAG,MAAM,IAAI,IAAI,CAAC;IACnC,IAAI,aAAa,GAAG,MAAM,IAAI,IAAI,CAAC;IAEnC,MAAM,KAAK,GAAG,MAAM,cAAc,CAAC;QAC/B,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC;QACpC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC;QACnC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,sBAAsB,CAAC;KAC3C,EAAE,KAAK,IAAI,EAAE,CAAC,IAAI,UAAU,CAAC;QAC1B,KAAK;QACL,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,QAAQ,CAAC;QAChD,SAAS,EAAE,SAAS,IAAI,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;KACvD,CAAC,CAAC,CAAC;IACJ,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC;QACjC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC;QACvC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,iBAAiB,CAAC;QACtC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC;KAC/C,EAAE,KAAK,IAAI,EAAE,CAAC,IAAI,YAAY,CAAC;QAC5B,KAAK;QACL,WAAW;QACX,SAAS;QACT,OAAO;KACV,CAAC,CAAC,CAAC;IAEJ,MAAM,UAAU,GAAG,IAAI,eAAe,CAAC;QACnC,eAAe,EAAE,OAAO,CAAC,WAAW,EAAE;KACzC,CAAC,CAAC;IAEH,MAAM,IAAI,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,mCAAmC;IAEzF,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC,IAAI,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC;IAExE,IAAI,YAAY;QACZ,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC,CAAC;IAExE,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,qBAAqB,CAAC,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC,CAAC;IACjF,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC;IACxE,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,iBAAiB,CAAC,IAAI,aAAa,mBAAmB,uBAAuB,UAAU,CAAC,CAAC;IAEtH,IAAI,sBAAsB,IAAI,IAAI;QAC9B,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,2BAA2B,CAAC,IAAI,sBAAsB,EAAE,CAAC,CAAC;IAE3F,IAAI,qBAAqB,IAAI,IAAI;QAC7B,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,0BAA0B,CAAC,IAAI,qBAAqB,EAAE,CAAC,CAAC;IAEzF,IAAI,CAAC,wBAAwB;QACzB,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,8BAA8B,CAAC,WAAW,CAAC,CAAC;IAE7E,iDAAiD;IACjD,MAAM,IAAI,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;IAErD,IAAI,CAAC,UAAU,CAAC,eAAe,EAAE;QAC7B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,wCAAwC,CAAC,CAAC,CAAC;QACjE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;KACnB;IAED,MAAM,iBAAiB,GAAa,EAAE,CAAC;IACvC,MAAM,iBAAiB,GAAa,EAAE,CAAC;IAEvC,KAAK,UAAU,QAAQ,CAAC,IAAyB;QAC7C,MAAM,EAAE,GAAG,QAAQ,CAAC,eAAe,CAAC;YAChC,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,OAAO,EAAE,IAAI,KAAK,QAAQ;gBACtB,CAAC,CAAC,iBAAiB,CAAC,KAAK,EAAE;gBAC3B,CAAC,CAAC,iBAAiB,CAAC,KAAK,EAAE;SAClC,CAAC,CAAC;QAEH,MAAM,GAAG,GAAW,MAAM,IAAI,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,GAAG,IAAI,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;QAClG,EAAE,CAAC,KAAK,EAAE,CAAC;QAEX,OAAO,GAAG,CAAC;IACf,CAAC;IAED,iDAAiD;IACjD,OAAO,IAAI,EAAE;QACT,MAAM,WAAW,GAAG,aAAa,IAAI,IAAI;YACrC,CAAC,CAAC,aAAa;YACf,CAAC,CAAC,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAE/B,IAAI,aAAa,IAAI,IAAI,EAAE;YACvB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,aAAa,CAAC,CAAC;YACrD,aAAa,GAAG,IAAI,CAAC;SACxB;;YACG,MAAM,iBAAiB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAE9C,IAAI,WAAW,KAAK,OAAO;YACvB,MAAM;QAEV,MAAM,WAAW,GAAG,aAAa,IAAI,IAAI;YACrC,CAAC,CAAC,aAAa;YACf,CAAC,CAAC,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAE/B,IAAI,aAAa,IAAI,IAAI,EAAE;YACvB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,aAAa,CAAC,CAAC;YACrD,aAAa,GAAG,IAAI,CAAC;SACxB;;YACG,MAAM,iBAAiB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAE9C,IAAI,WAAW,KAAK,OAAO;YACvB,MAAM;QAEV,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC;QAE/C,MAAM,CAAC,UAAU,EAAE,QAAQ,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QAEpE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;QACjC,MAAM,UAAU,CAAC,wBAAwB,CAAC,WAAW,EAAE,WAAW,EAAE;YAChE,WAAW;YACX,IAAI;YACJ,IAAI;YACJ,IAAI;YACJ,aAAa,EAAE;gBACX,OAAO,EAAE,aAAa;gBACtB,gBAAgB,EAAE,sBAAsB,IAAI,IAAI,CAAC,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,SAAS;gBACrF,eAAe,EAAE,qBAAqB,IAAI,IAAI,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,SAAS;gBAClF,eAAe,EAAE,wBAAwB;gBACzC,UAAU,EAAE,uBAAuB;aACtC;YACD,SAAS,EAAE,SAAS,KAAK,CAAC,CAAC;gBACvB,CAAC,CAAC,OAAO,CAAC,WAAW;gBACrB,CAAC,CAAC,SAAS,IAAI,CAAC;oBACZ,CAAC,CAAC,SAAS;oBACX,CAAC,CAAC,SAAS;YACnB,OAAO,CAAC,KAAK;gBACT,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC;YAClD,CAAC;SACJ,CAAC,CAAC;QACH,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QAC/B,OAAO,CAAC,GAAG,EAAE,CAAC;QAEd,IAAI,YAAY;YACZ,MAAM,OAAO,CAAC,YAAY,EAAE,CAAC;KACpC;AACL,CAAC"}
@@ -1,21 +1,18 @@
1
- import { defaultLlamaCppCudaSupport, defaultLlamaCppGitHubRepo, defaultLlamaCppMetalSupport, defaultLlamaCppRelease, defaultSkipDownload } from "../../config.js";
2
- import { getPrebuildBinPath } from "../../utils/getBin.js";
3
- import { DownloadLlamaCppCommand } from "./DownloadCommand.js";
1
+ import { defaultSkipDownload } from "../../config.js";
2
+ import { getLlamaForOptions } from "../../bindings/getLlama.js";
4
3
  export const OnPostInstallCommand = {
5
4
  command: "postinstall",
6
5
  describe: false,
7
6
  async handler() {
8
7
  if (defaultSkipDownload)
9
8
  return;
10
- if (await getPrebuildBinPath() != null)
11
- return;
12
9
  try {
13
- await DownloadLlamaCppCommand({
14
- repo: defaultLlamaCppGitHubRepo,
15
- release: defaultLlamaCppRelease,
16
- metal: defaultLlamaCppMetalSupport,
17
- cuda: defaultLlamaCppCudaSupport
10
+ await getLlamaForOptions({
11
+ progressLogs: true
12
+ }, {
13
+ updateLastBuildInfoOnCompile: true
18
14
  });
15
+ process.exit(0);
19
16
  }
20
17
  catch (err) {
21
18
  console.error(err);
@@ -1 +1 @@
1
- {"version":3,"file":"OnPostInstallCommand.js","sourceRoot":"","sources":["../../../src/cli/commands/OnPostInstallCommand.ts"],"names":[],"mappings":"AACA,OAAO,EACH,0BAA0B,EAAE,yBAAyB,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,mBAAmB,EAClI,MAAM,iBAAiB,CAAC;AACzB,OAAO,EAAC,kBAAkB,EAAC,MAAM,uBAAuB,CAAC;AACzD,OAAO,EAAC,uBAAuB,EAAC,MAAM,sBAAsB,CAAC;AAI7D,MAAM,CAAC,MAAM,oBAAoB,GAAgD;IAC7E,OAAO,EAAE,aAAa;IACtB,QAAQ,EAAE,KAAK;IACf,KAAK,CAAC,OAAO;QACT,IAAI,mBAAmB;YACnB,OAAO;QAEX,IAAI,MAAM,kBAAkB,EAAE,IAAI,IAAI;YAClC,OAAO;QAEX,IAAI;YACA,MAAM,uBAAuB,CAAC;gBAC1B,IAAI,EAAE,yBAAyB;gBAC/B,OAAO,EAAE,sBAAsB;gBAC/B,KAAK,EAAE,2BAA2B;gBAClC,IAAI,EAAE,0BAA0B;aACnC,CAAC,CAAC;SACN;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACnB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;SACnB;IACL,CAAC;CACJ,CAAC"}
1
+ {"version":3,"file":"OnPostInstallCommand.js","sourceRoot":"","sources":["../../../src/cli/commands/OnPostInstallCommand.ts"],"names":[],"mappings":"AACA,OAAO,EAAC,mBAAmB,EAAC,MAAM,iBAAiB,CAAC;AACpD,OAAO,EAAC,kBAAkB,EAAC,MAAM,4BAA4B,CAAC;AAI9D,MAAM,CAAC,MAAM,oBAAoB,GAAgD;IAC7E,OAAO,EAAE,aAAa;IACtB,QAAQ,EAAE,KAAK;IACf,KAAK,CAAC,OAAO;QACT,IAAI,mBAAmB;YACnB,OAAO;QAEX,IAAI;YACA,MAAM,kBAAkB,CAAC;gBACrB,YAAY,EAAE,IAAI;aACrB,EAAE;gBACC,4BAA4B,EAAE,IAAI;aACrC,CAAC,CAAC;YAEH,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;SACnB;QAAC,OAAO,GAAG,EAAE;YACV,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACnB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;SACnB;IACL,CAAC;CACJ,CAAC"}
package/dist/config.d.ts CHANGED
@@ -1,24 +1,31 @@
1
+ import { LlamaLogLevel } from "./bindings/types.js";
1
2
  export declare const llamaDirectory: string;
2
3
  export declare const llamaToolchainsDirectory: string;
3
- export declare const llamaBinsDirectory: string;
4
+ export declare const llamaPrebuiltBinsDirectory: string;
5
+ export declare const llamaLocalBuildBinsDirectory: string;
4
6
  export declare const llamaBinsGrammarsDirectory: string;
5
7
  export declare const llamaCppDirectory: string;
6
8
  export declare const llamaCppGrammarsDirectory: string;
7
9
  export declare const tempDownloadDirectory: string;
8
10
  export declare const chatCommandHistoryFilePath: string;
9
- export declare const usedBinFlagJsonPath: string;
11
+ export declare const lastBuildInfoJsonPath: string;
10
12
  export declare const binariesGithubReleasePath: string;
11
- export declare const llamaCppDirectoryTagFilePath: string;
13
+ export declare const llamaCppDirectoryInfoFilePath: string;
12
14
  export declare const currentReleaseGitBundlePath: string;
13
15
  export declare const xpackDirectory: string;
14
16
  export declare const localXpacksStoreDirectory: string;
15
17
  export declare const localXpacksCacheDirectory: string;
18
+ export declare const buildMetadataFileName = ".buildMetadata.json";
16
19
  export declare const xpmVersion = "^0.16.3";
20
+ export declare const builtinLlamaCppGitHubRepo = "ggerganov/llama.cpp";
21
+ export declare const builtinLlamaCppRelease: string;
17
22
  export declare const isCI: boolean;
23
+ export declare const isRunningInsideGoogleColab: boolean;
18
24
  export declare const defaultLlamaCppGitHubRepo: string;
19
25
  export declare const defaultLlamaCppRelease: string;
20
26
  export declare const defaultLlamaCppMetalSupport: boolean;
21
27
  export declare const defaultLlamaCppCudaSupport: boolean;
28
+ export declare const defaultLlamaCppDebugLogs: LlamaLogLevel;
22
29
  export declare const defaultSkipDownload: boolean;
23
30
  export declare const defaultXpacksStoreDirectory: string;
24
31
  export declare const defaultXpacksCacheDirectory: string;
package/dist/config.js CHANGED
@@ -4,40 +4,51 @@ import * as os from "os";
4
4
  import process from "process";
5
5
  import envVar from "env-var";
6
6
  import * as uuid from "uuid";
7
- import { getBinariesGithubRelease } from "./utils/binariesGithubRelease.js";
7
+ import { getBinariesGithubRelease } from "./bindings/utils/binariesGithubRelease.js";
8
+ import { LlamaLogLevel, LlamaLogLevelValues } from "./bindings/types.js";
8
9
  const __dirname = path.dirname(fileURLToPath(import.meta.url));
9
10
  const env = envVar.from(process.env);
10
11
  export const llamaDirectory = path.join(__dirname, "..", "llama");
11
12
  export const llamaToolchainsDirectory = path.join(llamaDirectory, "toolchains");
12
- export const llamaBinsDirectory = path.join(__dirname, "..", "llamaBins");
13
+ export const llamaPrebuiltBinsDirectory = path.join(__dirname, "..", "llamaBins");
14
+ export const llamaLocalBuildBinsDirectory = path.join(llamaDirectory, "localBuilds");
13
15
  export const llamaBinsGrammarsDirectory = path.join(__dirname, "..", "llama", "grammars");
14
16
  export const llamaCppDirectory = path.join(llamaDirectory, "llama.cpp");
15
17
  export const llamaCppGrammarsDirectory = path.join(llamaDirectory, "llama.cpp", "grammars");
16
18
  export const tempDownloadDirectory = path.join(os.tmpdir(), "node-llama-cpp", uuid.v4());
17
19
  export const chatCommandHistoryFilePath = path.join(os.homedir(), ".node-llama-cpp.chat_repl_history");
18
- export const usedBinFlagJsonPath = path.join(llamaDirectory, "usedBin.json");
20
+ export const lastBuildInfoJsonPath = path.join(llamaDirectory, "lastBuild.json");
19
21
  export const binariesGithubReleasePath = path.join(llamaDirectory, "binariesGithubRelease.json");
20
- export const llamaCppDirectoryTagFilePath = path.join(llamaDirectory, "llama.cpp.tag.json");
22
+ export const llamaCppDirectoryInfoFilePath = path.join(llamaDirectory, "llama.cpp.info.json");
21
23
  export const currentReleaseGitBundlePath = path.join(llamaDirectory, "gitRelease.bundle");
22
24
  export const xpackDirectory = path.join(llamaDirectory, "xpack");
23
25
  export const localXpacksStoreDirectory = path.join(xpackDirectory, "store");
24
26
  export const localXpacksCacheDirectory = path.join(xpackDirectory, "cache");
27
+ export const buildMetadataFileName = ".buildMetadata.json";
25
28
  export const xpmVersion = "^0.16.3";
29
+ export const builtinLlamaCppGitHubRepo = "ggerganov/llama.cpp";
30
+ export const builtinLlamaCppRelease = await getBinariesGithubRelease();
26
31
  export const isCI = env.get("CI")
27
32
  .default("false")
28
33
  .asBool();
34
+ export const isRunningInsideGoogleColab = env.get("COLAB_RELEASE_TAG")
35
+ .default("")
36
+ .asString() !== "";
29
37
  export const defaultLlamaCppGitHubRepo = env.get("NODE_LLAMA_CPP_REPO")
30
- .default("ggerganov/llama.cpp")
38
+ .default(builtinLlamaCppGitHubRepo)
31
39
  .asString();
32
40
  export const defaultLlamaCppRelease = env.get("NODE_LLAMA_CPP_REPO_RELEASE")
33
- .default(await getBinariesGithubRelease())
41
+ .default(builtinLlamaCppRelease)
34
42
  .asString();
35
43
  export const defaultLlamaCppMetalSupport = env.get("NODE_LLAMA_CPP_METAL")
36
- .default(process.platform === "darwin" ? "true" : "false")
44
+ .default((process.platform === "darwin" && process.arch !== "x64") ? "true" : "false")
37
45
  .asBool();
38
46
  export const defaultLlamaCppCudaSupport = env.get("NODE_LLAMA_CPP_CUDA")
39
47
  .default("false")
40
48
  .asBool();
49
+ export const defaultLlamaCppDebugLogs = env.get("NODE_LLAMA_CPP_LOG_LEVEL")
50
+ .default(LlamaLogLevel.debug)
51
+ .asEnum(LlamaLogLevelValues);
41
52
  export const defaultSkipDownload = env.get("NODE_LLAMA_CPP_SKIP_DOWNLOAD")
42
53
  .default("false")
43
54
  .asBool();
@@ -1 +1 @@
1
- {"version":3,"file":"config.js","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,MAAM,MAAM,SAAS,CAAC;AAC7B,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAC,wBAAwB,EAAC,MAAM,kCAAkC,CAAC;AAE1E,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;AAGrC,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;AAClE,MAAM,CAAC,MAAM,wBAAwB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,YAAY,CAAC,CAAC;AAChF,MAAM,CAAC,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC;AAC1E,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,gBAAgB,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,mCAAmC,CAAC,CAAC;AACvG,MAAM,CAAC,MAAM,mBAAmB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,cAAc,CAAC,CAAC;AAC7E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,4BAA4B,CAAC,CAAC;AACjG,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,oBAAoB,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AACjE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,UAAU,GAAG,SAAS,CAAC;AAEpC,MAAM,CAAC,MAAM,IAAI,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC;KAC5B,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KAClE,OAAO,CAAC,qBAAqB,CAAC;KAC9B,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,sBAAsB,GAAG,GAAG,CAAC,GAAG,CAAC,6BAA6B,CAAC;KACvE,OAAO,CAAC,MAAM,wBAAwB,EAAE,CAAC;KACzC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,sBAAsB,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;KACzD,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KACnE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,mBAAmB,GAAG,GAAG,CAAC,GAAG,CAAC,8BAA8B,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,8BAA8B,GAAG,8BAA8B,CAAC;AAC7E,MAAM,CAAC,MAAM,uBAAuB,GAAG,+FAA+F;IAClI,+HAA+H;IAC/H,mFAAmF,CAAC;AACxF,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAAC;AAC3C,MAAM,CAAC,MAAM,YAAY,GAAG,WAAW,CAAC;AAExC,MAAM,gBAAgB,GAAG,4CAA4C,CAAC;AACtE,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACjC,IAAI,EAAE,gBAAgB,GAAG,aAAa;CAChC,CAAC"}
1
+ {"version":3,"file":"config.js","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,aAAa,EAAC,MAAM,KAAK,CAAC;AAClC,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,MAAM,MAAM,SAAS,CAAC;AAC7B,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAC,wBAAwB,EAAC,MAAM,2CAA2C,CAAC;AACnF,OAAO,EAAC,aAAa,EAAE,mBAAmB,EAAC,MAAM,qBAAqB,CAAC;AAEvE,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE/D,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;AAGrC,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;AAClE,MAAM,CAAC,MAAM,wBAAwB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,YAAY,CAAC,CAAC;AAChF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC;AAClF,MAAM,CAAC,MAAM,4BAA4B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,aAAa,CAAC,CAAC;AACrF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,iBAAiB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;AACxE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC;AAC5F,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,gBAAgB,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,mCAAmC,CAAC,CAAC;AACvG,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,gBAAgB,CAAC,CAAC;AACjF,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,4BAA4B,CAAC,CAAC;AACjG,MAAM,CAAC,MAAM,6BAA6B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,qBAAqB,CAAC,CAAC;AAC9F,MAAM,CAAC,MAAM,2BAA2B,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,CAAC;AAC1F,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AACjE,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,OAAO,CAAC,CAAC;AAC5E,MAAM,CAAC,MAAM,qBAAqB,GAAG,qBAAqB,CAAC;AAC3D,MAAM,CAAC,MAAM,UAAU,GAAG,SAAS,CAAC;AACpC,MAAM,CAAC,MAAM,yBAAyB,GAAG,qBAAqB,CAAC;AAC/D,MAAM,CAAC,MAAM,sBAAsB,GAAG,MAAM,wBAAwB,EAAE,CAAC;AAEvE,MAAM,CAAC,MAAM,IAAI,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC;KAC5B,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC,GAAG,CAAC,mBAAmB,CAAC;KACjE,OAAO,CAAC,EAAE,CAAC;KACX,QAAQ,EAAE,KAAK,EAAE,CAAC;AACvB,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KAClE,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,sBAAsB,GAAG,GAAG,CAAC,GAAG,CAAC,6BAA6B,CAAC;KACvE,OAAO,CAAC,sBAAsB,CAAC;KAC/B,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,sBAAsB,CAAC;KACrE,OAAO,CAAC,CAAC,OAAO,CAAC,QAAQ,KAAK,QAAQ,IAAI,OAAO,CAAC,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;KACrF,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,0BAA0B,GAAG,GAAG,CAAC,GAAG,CAAC,qBAAqB,CAAC;KACnE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,wBAAwB,GAAG,GAAG,CAAC,GAAG,CAAC,0BAA0B,CAAC;KACtE,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC;KAC5B,MAAM,CAAC,mBAAmB,CAAC,CAAC;AACjC,MAAM,CAAC,MAAM,mBAAmB,GAAG,GAAG,CAAC,GAAG,CAAC,8BAA8B,CAAC;KACrE,OAAO,CAAC,OAAO,CAAC;KAChB,MAAM,EAAE,CAAC;AACd,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,2BAA2B,GAAG,GAAG,CAAC,GAAG,CAAC,oCAAoC,CAAC;KACnF,OAAO,CAAC,yBAAyB,CAAC;KAClC,QAAQ,EAAE,CAAC;AAChB,MAAM,CAAC,MAAM,8BAA8B,GAAG,8BAA8B,CAAC;AAC7E,MAAM,CAAC,MAAM,uBAAuB,GAAG,+FAA+F;IAClI,+HAA+H;IAC/H,mFAAmF,CAAC;AACxF,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAAC;AAC3C,MAAM,CAAC,MAAM,YAAY,GAAG,WAAW,CAAC;AAExC,MAAM,gBAAgB,GAAG,4CAA4C,CAAC;AACtE,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACjC,IAAI,EAAE,gBAAgB,GAAG,aAAa;CAChC,CAAC"}
@@ -0,0 +1 @@
1
+ export declare const UNKNOWN_UNICODE_CHAR = "\uFFFD";
package/dist/consts.js ADDED
@@ -0,0 +1,2 @@
1
+ export const UNKNOWN_UNICODE_CHAR = "\ufffd";
2
+ //# sourceMappingURL=consts.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"consts.js","sourceRoot":"","sources":["../src/consts.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,oBAAoB,GAAG,QAAQ,CAAC"}
@@ -0,0 +1,154 @@
1
+ /// <reference types="node" />
2
+ import { EventRelay } from "lifecycle-utils";
3
+ import { ChatWrapper } from "../../ChatWrapper.js";
4
+ import { LlamaContextSequence } from "../LlamaContext/LlamaContext.js";
5
+ import { ChatHistoryItem, ChatModelFunctions, LLamaContextualRepeatPenalty, Token } from "../../types.js";
6
+ import { GbnfJsonSchemaToType } from "../../utils/gbnfJson/types.js";
7
+ import { LlamaGrammar } from "../LlamaGrammar.js";
8
+ import { EvaluationPriority } from "../LlamaContext/types.js";
9
+ export type LlamaChatOptions = {
10
+ contextSequence: LlamaContextSequence;
11
+ /** `"auto"` is used by default */
12
+ chatWrapper?: "auto" | ChatWrapper;
13
+ /** Automatically dispose the sequence when the session is disposed */
14
+ autoDisposeSequence?: boolean;
15
+ };
16
+ export type LLamaChatGenerateResponseOptions<Functions extends ChatModelFunctions | undefined = undefined> = {
17
+ onToken?: (tokens: Token[]) => void;
18
+ signal?: AbortSignal;
19
+ maxTokens?: number;
20
+ /**
21
+ * Temperature is a hyperparameter that controls the randomness of the generated text.
22
+ * It affects the probability distribution of the model's output tokens.
23
+ * A higher temperature (e.g., 1.5) makes the output more random and creative,
24
+ * while a lower temperature (e.g., 0.5) makes the output more focused, deterministic, and conservative.
25
+ * The suggested temperature is 0.8, which provides a balance between randomness and determinism.
26
+ * At the extreme, a temperature of 0 will always pick the most likely next token, leading to identical outputs in each run.
27
+ *
28
+ * Set to `0` to disable.
29
+ * Disabled by default (set to `0`).
30
+ */
31
+ temperature?: number;
32
+ /**
33
+ * From the next token candidates, discard the percentage of tokens with the lowest probability.
34
+ * For example, if set to `0.05`, 5% of the lowest probability tokens will be discarded.
35
+ * This is useful for generating more high-quality results when using a high temperature.
36
+ * Set to a value between `0` and `1` to enable.
37
+ *
38
+ * Only relevant when `temperature` is set to a value greater than `0`.
39
+ * Disabled by default.
40
+ */
41
+ minP?: number;
42
+ /**
43
+ * Limits the model to consider only the K most likely next tokens for sampling at each step of sequence generation.
44
+ * An integer number between `1` and the size of the vocabulary.
45
+ * Set to `0` to disable (which uses the full vocabulary).
46
+ *
47
+ * Only relevant when `temperature` is set to a value greater than 0.
48
+ */
49
+ topK?: number;
50
+ /**
51
+ * Dynamically selects the smallest set of tokens whose cumulative probability exceeds the threshold P,
52
+ * and samples the next token only from this set.
53
+ * A float number between `0` and `1`.
54
+ * Set to `1` to disable.
55
+ *
56
+ * Only relevant when `temperature` is set to a value greater than `0`.
57
+ */
58
+ topP?: number;
59
+ /**
60
+ * Trim whitespace from the end of the generated text
61
+ * Disabled by default.
62
+ */
63
+ trimWhitespaceSuffix?: boolean;
64
+ repeatPenalty?: false | LLamaContextualRepeatPenalty;
65
+ /**
66
+ * See the parameter `evaluationPriority` on the `LlamaContextSequence.evaluate()` function for more information.
67
+ */
68
+ evaluationPriority?: EvaluationPriority;
69
+ contextShift?: LLamaChatContextShiftOptions;
70
+ /**
71
+ * The evaluation context window returned from the last evaluation.
72
+ * This is an optimization to utilize existing context sequence state better when possible.
73
+ */
74
+ lastEvaluationContextWindow?: {
75
+ /** The history of the last evaluation. */
76
+ history?: ChatHistoryItem[];
77
+ /**
78
+ * Minimum overlap percentage with existing context sequence state to use the last evaluation context window.
79
+ * If the last evaluation context window is not used, a new context will be generated based on the full history,
80
+ * which will decrease the likelihood of another context shift happening so soon.
81
+ */
82
+ minimumOverlapPercentageToPreventContextShift?: number;
83
+ };
84
+ } & ({
85
+ grammar?: LlamaGrammar;
86
+ functions?: never;
87
+ documentFunctionParams?: never;
88
+ } | {
89
+ grammar?: never;
90
+ functions?: Functions | ChatModelFunctions;
91
+ documentFunctionParams?: boolean;
92
+ });
93
+ export type LLamaChatContextShiftOptions = {
94
+ /**
95
+ * The number of tokens to delete from the context window to make space for new ones.
96
+ * Defaults to 10% of the context size.
97
+ */
98
+ size?: number | ((sequence: LlamaContextSequence) => number | Promise<number>);
99
+ /**
100
+ * The strategy to use when deleting tokens from the context window.
101
+ * Defaults to `"eraseFirstResponseAndKeepFirstSystem"`.
102
+ */
103
+ strategy?: "eraseFirstResponseAndKeepFirstSystem" | ((options: {
104
+ chatHistory: ChatHistoryItem[];
105
+ maxTokensCount: number;
106
+ tokenizer(text: string, specialTokens?: boolean): Token[];
107
+ chatWrapper: ChatWrapper;
108
+ lastShiftMetadata?: object | null;
109
+ }) => {
110
+ chatHistory: ChatHistoryItem[];
111
+ metadata?: object | null;
112
+ } | Promise<{
113
+ chatHistory: ChatHistoryItem[];
114
+ metadata?: object | null;
115
+ }>);
116
+ /**
117
+ * The `contextShiftMetadata` returned from the last evaluation.
118
+ * This is an optimization to utilize the existing context state better when possible.
119
+ */
120
+ lastEvaluationMetadata?: object | undefined | null;
121
+ };
122
+ export declare class LlamaChat {
123
+ readonly onDispose: EventRelay<void>;
124
+ constructor({ contextSequence, chatWrapper, autoDisposeSequence }: LlamaChatOptions);
125
+ dispose({ disposeSequence }?: {
126
+ disposeSequence?: boolean;
127
+ }): void;
128
+ /** @hidden */
129
+ [Symbol.dispose](): void;
130
+ get disposed(): boolean;
131
+ get chatWrapper(): ChatWrapper;
132
+ get sequence(): LlamaContextSequence;
133
+ get context(): import("../LlamaContext/LlamaContext.js").LlamaContext;
134
+ get model(): import("../LlamaModel.js").LlamaModel;
135
+ generateResponse<const Functions extends ChatModelFunctions | undefined = undefined>(history: ChatHistoryItem[], { onToken, signal, maxTokens, temperature, minP, topK, topP, grammar, trimWhitespaceSuffix, repeatPenalty, evaluationPriority, functions, documentFunctionParams, contextShift, lastEvaluationContextWindow: { history: lastEvaluationContextWindowHistory, minimumOverlapPercentageToPreventContextShift } }?: LLamaChatGenerateResponseOptions<Functions>): Promise<LlamaChatResponse<Functions>>;
136
+ }
137
+ export type LlamaChatResponse<Functions extends ChatModelFunctions | undefined = undefined> = {
138
+ response: string;
139
+ functionCall?: Functions extends ChatModelFunctions ? LlamaChatResponseFunctionCall<Functions> : never;
140
+ lastEvaluation: {
141
+ cleanHistory: ChatHistoryItem[];
142
+ contextWindow: ChatHistoryItem[];
143
+ contextShiftMetadata: any;
144
+ };
145
+ metadata: {
146
+ remainingGenerationAfterStop?: string | Token[];
147
+ stopReason: "eosToken" | "stopGenerationTrigger" | "functionCall" | "maxTokens";
148
+ };
149
+ };
150
+ export type LlamaChatResponseFunctionCall<Functions extends ChatModelFunctions, FunctionCallName extends keyof Functions & string = string & keyof Functions, Params = Functions[FunctionCallName]["params"] extends undefined | null | void ? undefined : GbnfJsonSchemaToType<Functions[FunctionCallName]["params"]>> = {
151
+ functionName: FunctionCallName;
152
+ params: Params;
153
+ raw: string;
154
+ };