inference-server 1.0.0-beta.19

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 (227) hide show
  1. package/README.md +216 -0
  2. package/dist/api/openai/enums.d.ts +4 -0
  3. package/dist/api/openai/enums.js +17 -0
  4. package/dist/api/openai/enums.js.map +1 -0
  5. package/dist/api/openai/handlers/chat.d.ts +3 -0
  6. package/dist/api/openai/handlers/chat.js +358 -0
  7. package/dist/api/openai/handlers/chat.js.map +1 -0
  8. package/dist/api/openai/handlers/completions.d.ts +3 -0
  9. package/dist/api/openai/handlers/completions.js +169 -0
  10. package/dist/api/openai/handlers/completions.js.map +1 -0
  11. package/dist/api/openai/handlers/embeddings.d.ts +3 -0
  12. package/dist/api/openai/handlers/embeddings.js +74 -0
  13. package/dist/api/openai/handlers/embeddings.js.map +1 -0
  14. package/dist/api/openai/handlers/images.d.ts +0 -0
  15. package/dist/api/openai/handlers/images.js +4 -0
  16. package/dist/api/openai/handlers/images.js.map +1 -0
  17. package/dist/api/openai/handlers/models.d.ts +3 -0
  18. package/dist/api/openai/handlers/models.js +23 -0
  19. package/dist/api/openai/handlers/models.js.map +1 -0
  20. package/dist/api/openai/handlers/transcription.d.ts +0 -0
  21. package/dist/api/openai/handlers/transcription.js +4 -0
  22. package/dist/api/openai/handlers/transcription.js.map +1 -0
  23. package/dist/api/openai/index.d.ts +7 -0
  24. package/dist/api/openai/index.js +14 -0
  25. package/dist/api/openai/index.js.map +1 -0
  26. package/dist/api/parseJSONRequestBody.d.ts +2 -0
  27. package/dist/api/parseJSONRequestBody.js +24 -0
  28. package/dist/api/parseJSONRequestBody.js.map +1 -0
  29. package/dist/api/v1/index.d.ts +2 -0
  30. package/dist/api/v1/index.js +29 -0
  31. package/dist/api/v1/index.js.map +1 -0
  32. package/dist/cli.d.ts +1 -0
  33. package/dist/cli.js +10 -0
  34. package/dist/cli.js.map +1 -0
  35. package/dist/engines/gpt4all/engine.d.ts +34 -0
  36. package/dist/engines/gpt4all/engine.js +357 -0
  37. package/dist/engines/gpt4all/engine.js.map +1 -0
  38. package/dist/engines/gpt4all/util.d.ts +3 -0
  39. package/dist/engines/gpt4all/util.js +29 -0
  40. package/dist/engines/gpt4all/util.js.map +1 -0
  41. package/dist/engines/index.d.ts +19 -0
  42. package/dist/engines/index.js +21 -0
  43. package/dist/engines/index.js.map +1 -0
  44. package/dist/engines/node-llama-cpp/engine.d.ts +49 -0
  45. package/dist/engines/node-llama-cpp/engine.js +666 -0
  46. package/dist/engines/node-llama-cpp/engine.js.map +1 -0
  47. package/dist/engines/node-llama-cpp/types.d.ts +13 -0
  48. package/dist/engines/node-llama-cpp/types.js +2 -0
  49. package/dist/engines/node-llama-cpp/types.js.map +1 -0
  50. package/dist/engines/node-llama-cpp/util.d.ts +15 -0
  51. package/dist/engines/node-llama-cpp/util.js +84 -0
  52. package/dist/engines/node-llama-cpp/util.js.map +1 -0
  53. package/dist/engines/node-llama-cpp/validateModelFile.d.ts +8 -0
  54. package/dist/engines/node-llama-cpp/validateModelFile.js +36 -0
  55. package/dist/engines/node-llama-cpp/validateModelFile.js.map +1 -0
  56. package/dist/engines/stable-diffusion-cpp/engine.d.ts +90 -0
  57. package/dist/engines/stable-diffusion-cpp/engine.js +294 -0
  58. package/dist/engines/stable-diffusion-cpp/engine.js.map +1 -0
  59. package/dist/engines/stable-diffusion-cpp/types.d.ts +3 -0
  60. package/dist/engines/stable-diffusion-cpp/types.js +2 -0
  61. package/dist/engines/stable-diffusion-cpp/types.js.map +1 -0
  62. package/dist/engines/stable-diffusion-cpp/util.d.ts +4 -0
  63. package/dist/engines/stable-diffusion-cpp/util.js +55 -0
  64. package/dist/engines/stable-diffusion-cpp/util.js.map +1 -0
  65. package/dist/engines/stable-diffusion-cpp/validateModelFiles.d.ts +19 -0
  66. package/dist/engines/stable-diffusion-cpp/validateModelFiles.js +91 -0
  67. package/dist/engines/stable-diffusion-cpp/validateModelFiles.js.map +1 -0
  68. package/dist/engines/transformers-js/engine.d.ts +37 -0
  69. package/dist/engines/transformers-js/engine.js +538 -0
  70. package/dist/engines/transformers-js/engine.js.map +1 -0
  71. package/dist/engines/transformers-js/types.d.ts +7 -0
  72. package/dist/engines/transformers-js/types.js +2 -0
  73. package/dist/engines/transformers-js/types.js.map +1 -0
  74. package/dist/engines/transformers-js/util.d.ts +7 -0
  75. package/dist/engines/transformers-js/util.js +36 -0
  76. package/dist/engines/transformers-js/util.js.map +1 -0
  77. package/dist/engines/transformers-js/validateModelFiles.d.ts +17 -0
  78. package/dist/engines/transformers-js/validateModelFiles.js +133 -0
  79. package/dist/engines/transformers-js/validateModelFiles.js.map +1 -0
  80. package/dist/experiments/ChatWithVision.d.ts +11 -0
  81. package/dist/experiments/ChatWithVision.js +91 -0
  82. package/dist/experiments/ChatWithVision.js.map +1 -0
  83. package/dist/experiments/StableDiffPromptGenerator.d.ts +0 -0
  84. package/dist/experiments/StableDiffPromptGenerator.js +4 -0
  85. package/dist/experiments/StableDiffPromptGenerator.js.map +1 -0
  86. package/dist/experiments/VoiceFunctionCall.d.ts +18 -0
  87. package/dist/experiments/VoiceFunctionCall.js +51 -0
  88. package/dist/experiments/VoiceFunctionCall.js.map +1 -0
  89. package/dist/http.d.ts +19 -0
  90. package/dist/http.js +54 -0
  91. package/dist/http.js.map +1 -0
  92. package/dist/index.d.ts +7 -0
  93. package/dist/index.js +8 -0
  94. package/dist/index.js.map +1 -0
  95. package/dist/instance.d.ts +88 -0
  96. package/dist/instance.js +594 -0
  97. package/dist/instance.js.map +1 -0
  98. package/dist/lib/acquireFileLock.d.ts +7 -0
  99. package/dist/lib/acquireFileLock.js +38 -0
  100. package/dist/lib/acquireFileLock.js.map +1 -0
  101. package/dist/lib/calculateContextIdentity.d.ts +7 -0
  102. package/dist/lib/calculateContextIdentity.js +39 -0
  103. package/dist/lib/calculateContextIdentity.js.map +1 -0
  104. package/dist/lib/calculateFileChecksum.d.ts +1 -0
  105. package/dist/lib/calculateFileChecksum.js +16 -0
  106. package/dist/lib/calculateFileChecksum.js.map +1 -0
  107. package/dist/lib/copyDirectory.d.ts +6 -0
  108. package/dist/lib/copyDirectory.js +27 -0
  109. package/dist/lib/copyDirectory.js.map +1 -0
  110. package/dist/lib/decodeAudio.d.ts +1 -0
  111. package/dist/lib/decodeAudio.js +26 -0
  112. package/dist/lib/decodeAudio.js.map +1 -0
  113. package/dist/lib/downloadModelFile.d.ts +10 -0
  114. package/dist/lib/downloadModelFile.js +58 -0
  115. package/dist/lib/downloadModelFile.js.map +1 -0
  116. package/dist/lib/flattenMessageTextContent.d.ts +2 -0
  117. package/dist/lib/flattenMessageTextContent.js +11 -0
  118. package/dist/lib/flattenMessageTextContent.js.map +1 -0
  119. package/dist/lib/getCacheDirPath.d.ts +12 -0
  120. package/dist/lib/getCacheDirPath.js +31 -0
  121. package/dist/lib/getCacheDirPath.js.map +1 -0
  122. package/dist/lib/loadImage.d.ts +12 -0
  123. package/dist/lib/loadImage.js +30 -0
  124. package/dist/lib/loadImage.js.map +1 -0
  125. package/dist/lib/logger.d.ts +12 -0
  126. package/dist/lib/logger.js +98 -0
  127. package/dist/lib/logger.js.map +1 -0
  128. package/dist/lib/math.d.ts +7 -0
  129. package/dist/lib/math.js +30 -0
  130. package/dist/lib/math.js.map +1 -0
  131. package/dist/lib/resolveModelFileLocation.d.ts +15 -0
  132. package/dist/lib/resolveModelFileLocation.js +41 -0
  133. package/dist/lib/resolveModelFileLocation.js.map +1 -0
  134. package/dist/lib/util.d.ts +7 -0
  135. package/dist/lib/util.js +61 -0
  136. package/dist/lib/util.js.map +1 -0
  137. package/dist/lib/validateModelFile.d.ts +9 -0
  138. package/dist/lib/validateModelFile.js +62 -0
  139. package/dist/lib/validateModelFile.js.map +1 -0
  140. package/dist/lib/validateModelOptions.d.ts +3 -0
  141. package/dist/lib/validateModelOptions.js +23 -0
  142. package/dist/lib/validateModelOptions.js.map +1 -0
  143. package/dist/pool.d.ts +61 -0
  144. package/dist/pool.js +512 -0
  145. package/dist/pool.js.map +1 -0
  146. package/dist/server.d.ts +59 -0
  147. package/dist/server.js +221 -0
  148. package/dist/server.js.map +1 -0
  149. package/dist/standalone.d.ts +1 -0
  150. package/dist/standalone.js +306 -0
  151. package/dist/standalone.js.map +1 -0
  152. package/dist/store.d.ts +60 -0
  153. package/dist/store.js +203 -0
  154. package/dist/store.js.map +1 -0
  155. package/dist/types/completions.d.ts +57 -0
  156. package/dist/types/completions.js +2 -0
  157. package/dist/types/completions.js.map +1 -0
  158. package/dist/types/index.d.ts +326 -0
  159. package/dist/types/index.js +2 -0
  160. package/dist/types/index.js.map +1 -0
  161. package/docs/engines.md +28 -0
  162. package/docs/gpu.md +72 -0
  163. package/docs/http-api.md +147 -0
  164. package/examples/all-options.js +108 -0
  165. package/examples/chat-cli.js +56 -0
  166. package/examples/chat-server.js +65 -0
  167. package/examples/concurrency.js +70 -0
  168. package/examples/express.js +70 -0
  169. package/examples/pool.js +91 -0
  170. package/package.json +113 -0
  171. package/src/api/openai/enums.ts +20 -0
  172. package/src/api/openai/handlers/chat.ts +408 -0
  173. package/src/api/openai/handlers/completions.ts +196 -0
  174. package/src/api/openai/handlers/embeddings.ts +92 -0
  175. package/src/api/openai/handlers/images.ts +3 -0
  176. package/src/api/openai/handlers/models.ts +33 -0
  177. package/src/api/openai/handlers/transcription.ts +2 -0
  178. package/src/api/openai/index.ts +16 -0
  179. package/src/api/parseJSONRequestBody.ts +26 -0
  180. package/src/api/v1/DRAFT.md +16 -0
  181. package/src/api/v1/index.ts +37 -0
  182. package/src/cli.ts +9 -0
  183. package/src/engines/gpt4all/engine.ts +441 -0
  184. package/src/engines/gpt4all/util.ts +31 -0
  185. package/src/engines/index.ts +28 -0
  186. package/src/engines/node-llama-cpp/engine.ts +811 -0
  187. package/src/engines/node-llama-cpp/types.ts +17 -0
  188. package/src/engines/node-llama-cpp/util.ts +126 -0
  189. package/src/engines/node-llama-cpp/validateModelFile.ts +46 -0
  190. package/src/engines/stable-diffusion-cpp/engine.ts +369 -0
  191. package/src/engines/stable-diffusion-cpp/types.ts +54 -0
  192. package/src/engines/stable-diffusion-cpp/util.ts +58 -0
  193. package/src/engines/stable-diffusion-cpp/validateModelFiles.ts +119 -0
  194. package/src/engines/transformers-js/engine.ts +659 -0
  195. package/src/engines/transformers-js/types.ts +25 -0
  196. package/src/engines/transformers-js/util.ts +40 -0
  197. package/src/engines/transformers-js/validateModelFiles.ts +168 -0
  198. package/src/experiments/ChatWithVision.ts +103 -0
  199. package/src/experiments/StableDiffPromptGenerator.ts +2 -0
  200. package/src/experiments/VoiceFunctionCall.ts +71 -0
  201. package/src/http.ts +72 -0
  202. package/src/index.ts +7 -0
  203. package/src/instance.ts +723 -0
  204. package/src/lib/acquireFileLock.ts +38 -0
  205. package/src/lib/calculateContextIdentity.ts +53 -0
  206. package/src/lib/calculateFileChecksum.ts +18 -0
  207. package/src/lib/copyDirectory.ts +29 -0
  208. package/src/lib/decodeAudio.ts +39 -0
  209. package/src/lib/downloadModelFile.ts +70 -0
  210. package/src/lib/flattenMessageTextContent.ts +19 -0
  211. package/src/lib/getCacheDirPath.ts +34 -0
  212. package/src/lib/loadImage.ts +46 -0
  213. package/src/lib/logger.ts +112 -0
  214. package/src/lib/math.ts +31 -0
  215. package/src/lib/resolveModelFileLocation.ts +49 -0
  216. package/src/lib/util.ts +75 -0
  217. package/src/lib/validateModelFile.ts +71 -0
  218. package/src/lib/validateModelOptions.ts +31 -0
  219. package/src/pool.ts +651 -0
  220. package/src/server.ts +270 -0
  221. package/src/standalone.ts +320 -0
  222. package/src/store.ts +278 -0
  223. package/src/types/completions.ts +86 -0
  224. package/src/types/index.ts +488 -0
  225. package/tsconfig.json +29 -0
  226. package/tsconfig.release.json +11 -0
  227. package/vitest.config.ts +18 -0
@@ -0,0 +1,357 @@
1
+ import path from 'node:path';
2
+ import fs from 'node:fs';
3
+ import { loadModel, createCompletion, createEmbedding, DEFAULT_MODEL_LIST_URL, } from 'gpt4all';
4
+ import { LogLevels } from '../../lib/logger.js';
5
+ import { downloadModelFile } from '../../lib/downloadModelFile.js';
6
+ import { acquireFileLock } from '../../lib/acquireFileLock.js';
7
+ import { validateModelFile } from '../../lib/validateModelFile.js';
8
+ import { createChatMessageArray } from './util.js';
9
+ export const autoGpu = true;
10
+ export async function prepareModel({ config, log }, onProgress, signal) {
11
+ fs.mkdirSync(path.dirname(config.location), { recursive: true });
12
+ const releaseFileLock = await acquireFileLock(config.location);
13
+ if (signal?.aborted) {
14
+ releaseFileLock();
15
+ return;
16
+ }
17
+ log(LogLevels.info, `Preparing gpt4all model at ${config.location}`, {
18
+ model: config.id,
19
+ });
20
+ let modelMeta;
21
+ let modelList;
22
+ const modelMetaPath = path.join(path.dirname(config.location), 'models.json');
23
+ try {
24
+ if (!fs.existsSync(modelMetaPath)) {
25
+ const res = await fetch(DEFAULT_MODEL_LIST_URL);
26
+ modelList = (await res.json());
27
+ fs.writeFileSync(modelMetaPath, JSON.stringify(modelList, null, 2));
28
+ }
29
+ else {
30
+ modelList = JSON.parse(fs.readFileSync(modelMetaPath, 'utf-8'));
31
+ }
32
+ const foundModelMeta = modelList.find((item) => {
33
+ if (config.md5 && item.md5sum) {
34
+ return item.md5sum === config.md5;
35
+ }
36
+ if (config.url && item.url) {
37
+ return item.url === config.url;
38
+ }
39
+ return item.filename === path.basename(config.location);
40
+ });
41
+ if (foundModelMeta) {
42
+ modelMeta = foundModelMeta;
43
+ }
44
+ const validationError = await validateModelFile({
45
+ ...config,
46
+ md5: config.md5 || modelMeta?.md5sum,
47
+ });
48
+ if (signal?.aborted) {
49
+ return;
50
+ }
51
+ if (validationError) {
52
+ if (config.url) {
53
+ log(LogLevels.info, 'Downloading', {
54
+ model: config.id,
55
+ url: config.url,
56
+ location: config.location,
57
+ error: validationError,
58
+ });
59
+ await downloadModelFile({
60
+ url: config.url,
61
+ filePath: config.location,
62
+ modelsCachePath: config.modelsCachePath,
63
+ onProgress,
64
+ signal,
65
+ });
66
+ }
67
+ else {
68
+ throw new Error(`${validationError} - No URL provided`);
69
+ }
70
+ }
71
+ const finalValidationError = await validateModelFile({
72
+ ...config,
73
+ md5: config.md5 || modelMeta?.md5sum,
74
+ });
75
+ if (finalValidationError) {
76
+ throw new Error(`Downloaded files are invalid: ${finalValidationError}`);
77
+ }
78
+ if (signal?.aborted) {
79
+ return;
80
+ }
81
+ return modelMeta;
82
+ }
83
+ catch (error) {
84
+ throw error;
85
+ }
86
+ finally {
87
+ releaseFileLock();
88
+ }
89
+ }
90
+ export async function createInstance({ config, log }, signal) {
91
+ log(LogLevels.info, `Load GPT4All model ${config.location}`);
92
+ let device = config.device?.gpu ?? 'cpu';
93
+ if (typeof device === 'boolean') {
94
+ device = device ? 'gpu' : 'cpu';
95
+ }
96
+ else if (device === 'auto') {
97
+ device = 'cpu';
98
+ }
99
+ const loadOpts = {
100
+ modelPath: path.dirname(config.location),
101
+ // file: config.file,
102
+ modelConfigFile: path.dirname(config.location) + '/models.json',
103
+ allowDownload: false,
104
+ device: device,
105
+ ngl: config.device?.gpuLayers ?? 100,
106
+ nCtx: config.contextSize ?? 2048,
107
+ // verbose: true,
108
+ // signal?: // TODO no way to cancel load
109
+ };
110
+ let modelType;
111
+ if (config.task === 'text-completion') {
112
+ modelType = 'inference';
113
+ }
114
+ else if (config.task === 'embedding') {
115
+ modelType = 'embedding';
116
+ }
117
+ else {
118
+ throw new Error(`Unsupported task type: ${config.task}`);
119
+ }
120
+ const instance = await loadModel(path.basename(config.location), {
121
+ ...loadOpts,
122
+ type: modelType,
123
+ });
124
+ if (config.device?.cpuThreads) {
125
+ instance.llm.setThreadCount(config.device.cpuThreads);
126
+ }
127
+ if ('generate' in instance) {
128
+ if (config.initialMessages?.length) {
129
+ let messages = createChatMessageArray(config.initialMessages);
130
+ let systemPrompt;
131
+ if (messages[0].role === 'system') {
132
+ systemPrompt = messages[0].content;
133
+ messages = messages.slice(1);
134
+ }
135
+ await instance.createChatSession({
136
+ systemPrompt,
137
+ messages,
138
+ });
139
+ }
140
+ else if (config.prefix) {
141
+ await instance.generate(config.prefix, {
142
+ nPredict: 0,
143
+ });
144
+ }
145
+ else {
146
+ await instance.generate('', {
147
+ nPredict: 0,
148
+ });
149
+ }
150
+ }
151
+ return instance;
152
+ }
153
+ export async function disposeInstance(instance) {
154
+ instance.dispose();
155
+ }
156
+ export async function processTextCompletionTask({ request, config, onChunk }, instance, signal) {
157
+ if (!('generate' in instance)) {
158
+ throw new Error('Instance does not support text completion.');
159
+ }
160
+ if (!request.prompt) {
161
+ throw new Error('Prompt is required for text completion.');
162
+ }
163
+ let finishReason = 'eogToken';
164
+ let suffixToRemove;
165
+ const defaults = config.completionDefaults ?? {};
166
+ const stopTriggers = request.stop ?? defaults.stop ?? [];
167
+ const includesStopTriggers = (text) => stopTriggers.find((t) => text.includes(t));
168
+ const result = await instance.generate(request.prompt, {
169
+ // @ts-ignore
170
+ special: true, // allows passing in raw prompt (including <|start|> etc.)
171
+ promptTemplate: '%1',
172
+ temperature: request.temperature ?? defaults.temperature,
173
+ nPredict: request.maxTokens ?? defaults.maxTokens,
174
+ topP: request.topP ?? defaults.topP,
175
+ topK: request.topK ?? defaults.topK,
176
+ minP: request.minP ?? defaults.minP,
177
+ nBatch: config?.batchSize,
178
+ repeatLastN: request.repeatPenaltyNum ?? defaults.repeatPenaltyNum,
179
+ // repeat penalty is doing something different than both frequency and presence penalty
180
+ // so not falling back to them here.
181
+ repeatPenalty: request.repeatPenalty ?? defaults.repeatPenalty,
182
+ // seed: args.seed, // https://github.com/nomic-ai/gpt4all/issues/1952
183
+ // @ts-ignore
184
+ onResponseToken: (tokenId, text) => {
185
+ const matchingTrigger = includesStopTriggers(text);
186
+ if (matchingTrigger) {
187
+ finishReason = 'stopTrigger';
188
+ suffixToRemove = text;
189
+ return false;
190
+ }
191
+ if (onChunk) {
192
+ onChunk({
193
+ text,
194
+ tokens: [tokenId],
195
+ });
196
+ }
197
+ return !signal?.aborted;
198
+ },
199
+ // @ts-ignore
200
+ onResponseTokens: ({ tokenIds, text }) => {
201
+ const matchingTrigger = includesStopTriggers(text);
202
+ if (matchingTrigger) {
203
+ finishReason = 'stopTrigger';
204
+ suffixToRemove = text;
205
+ return false;
206
+ }
207
+ if (onChunk) {
208
+ onChunk({
209
+ text,
210
+ tokens: tokenIds,
211
+ });
212
+ }
213
+ return !signal?.aborted;
214
+ },
215
+ });
216
+ if (result.tokensGenerated === request.maxTokens) {
217
+ finishReason = 'maxTokens';
218
+ }
219
+ let responseText = result.text;
220
+ if (suffixToRemove) {
221
+ responseText = responseText.slice(0, -suffixToRemove.length);
222
+ }
223
+ return {
224
+ finishReason,
225
+ text: responseText,
226
+ promptTokens: result.tokensIngested,
227
+ completionTokens: result.tokensGenerated,
228
+ contextTokens: instance.activeChatSession?.promptContext.nPast ?? 0,
229
+ };
230
+ }
231
+ export async function processChatCompletionTask({ request, config, resetContext, log, onChunk }, instance, signal) {
232
+ if (!('createChatSession' in instance)) {
233
+ throw new Error('Instance does not support chat completion.');
234
+ }
235
+ let session = instance.activeChatSession;
236
+ if (!session || resetContext) {
237
+ log(LogLevels.debug, 'Resetting chat context');
238
+ let messages = createChatMessageArray(request.messages);
239
+ let systemPrompt;
240
+ if (messages[0].role === 'system') {
241
+ systemPrompt = messages[0].content;
242
+ messages = messages.slice(1);
243
+ }
244
+ // drop last user message
245
+ if (messages[messages.length - 1].role === 'user') {
246
+ messages = messages.slice(0, -1);
247
+ }
248
+ session = await instance.createChatSession({
249
+ systemPrompt,
250
+ messages,
251
+ });
252
+ }
253
+ const conversationMessages = createChatMessageArray(request.messages).filter((m) => m.role !== 'system');
254
+ const lastMessage = conversationMessages[conversationMessages.length - 1];
255
+ if (!(lastMessage.role === 'user' && lastMessage.content)) {
256
+ throw new Error('Chat completions require a final user message.');
257
+ }
258
+ const input = lastMessage.content;
259
+ let finishReason = 'eogToken';
260
+ let suffixToRemove;
261
+ const defaults = config.completionDefaults ?? {};
262
+ const stopTriggers = request.stop ?? defaults.stop ?? [];
263
+ const includesStopTriggers = (text) => stopTriggers.find((t) => text.includes(t));
264
+ const result = await createCompletion(session, input, {
265
+ temperature: request.temperature ?? defaults.temperature,
266
+ nPredict: request.maxTokens ?? defaults.maxTokens,
267
+ topP: request.topP ?? defaults.topP,
268
+ topK: request.topK ?? defaults.topK,
269
+ minP: request.minP ?? defaults.minP,
270
+ nBatch: config.batchSize,
271
+ repeatLastN: request.repeatPenaltyNum ?? defaults.repeatPenaltyNum,
272
+ repeatPenalty: request.repeatPenalty ?? defaults.repeatPenalty,
273
+ // seed: args.seed, // see https://github.com/nomic-ai/gpt4all/issues/1952
274
+ // @ts-ignore
275
+ onResponseToken: (tokenId, text) => {
276
+ const matchingTrigger = includesStopTriggers(text);
277
+ if (matchingTrigger) {
278
+ finishReason = 'stopTrigger';
279
+ suffixToRemove = text;
280
+ return false;
281
+ }
282
+ if (onChunk) {
283
+ onChunk({
284
+ text,
285
+ tokens: [tokenId],
286
+ });
287
+ }
288
+ return !signal?.aborted;
289
+ },
290
+ // @ts-ignore
291
+ onResponseTokens: ({ tokenIds, text }) => {
292
+ const matchingTrigger = includesStopTriggers(text);
293
+ if (matchingTrigger) {
294
+ finishReason = 'stopTrigger';
295
+ suffixToRemove = text;
296
+ return false;
297
+ }
298
+ if (onChunk) {
299
+ onChunk({
300
+ tokens: tokenIds,
301
+ text,
302
+ });
303
+ }
304
+ return !signal?.aborted;
305
+ },
306
+ });
307
+ if (result.usage.completion_tokens === request.maxTokens) {
308
+ finishReason = 'maxTokens';
309
+ }
310
+ let response = result.choices[0].message.content;
311
+ if (suffixToRemove) {
312
+ response = response.slice(0, -suffixToRemove.length);
313
+ }
314
+ return {
315
+ finishReason,
316
+ message: {
317
+ role: 'assistant',
318
+ content: response,
319
+ },
320
+ promptTokens: result.usage.prompt_tokens,
321
+ completionTokens: result.usage.completion_tokens,
322
+ contextTokens: session.promptContext.nPast,
323
+ };
324
+ }
325
+ export async function processEmbeddingTask({ request, config }, instance, signal) {
326
+ if (!('embed' in instance)) {
327
+ throw new Error('Instance does not support embedding.');
328
+ }
329
+ if (!request.input) {
330
+ throw new Error('Input is required for embedding.');
331
+ }
332
+ const texts = [];
333
+ if (typeof request.input === 'string') {
334
+ texts.push(request.input);
335
+ }
336
+ else if (Array.isArray(request.input)) {
337
+ for (const input of request.input) {
338
+ if (typeof input === 'string') {
339
+ texts.push(input);
340
+ }
341
+ else if (input.type === 'text') {
342
+ texts.push(input.content);
343
+ }
344
+ else if (input.type === 'image') {
345
+ throw new Error('Image inputs not implemented.');
346
+ }
347
+ }
348
+ }
349
+ const res = await createEmbedding(instance, texts, {
350
+ dimensionality: request.dimensions,
351
+ });
352
+ return {
353
+ embeddings: res.embeddings,
354
+ inputTokens: res.n_prompt_tokens,
355
+ };
356
+ }
357
+ //# sourceMappingURL=engine.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"engine.js","sourceRoot":"","sources":["../../../src/engines/gpt4all/engine.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAA;AAC5B,OAAO,EAAE,MAAM,SAAS,CAAA;AACxB,OAAO,EACN,SAAS,EACT,gBAAgB,EAChB,eAAe,EAKf,sBAAsB,GACtB,MAAM,SAAS,CAAA;AAehB,OAAO,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAA;AAClD,OAAO,EAAE,iBAAiB,EAAE,MAAM,mCAAmC,CAAA;AACrE,OAAO,EAAE,eAAe,EAAE,MAAM,iCAAiC,CAAA;AACjE,OAAO,EAAE,iBAAiB,EAAE,MAAM,mCAAmC,CAAA;AACrE,OAAO,EAAE,sBAAsB,EAAE,MAAM,WAAW,CAAA;AA8BlD,MAAM,CAAC,MAAM,OAAO,GAAG,IAAI,CAAA;AAE3B,MAAM,CAAC,KAAK,UAAU,YAAY,CACjC,EAAE,MAAM,EAAE,GAAG,EAAqC,EAClD,UAAqD,EACrD,MAAoB;IAEpB,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAA;IAChE,MAAM,eAAe,GAAG,MAAM,eAAe,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAA;IAC9D,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;QACrB,eAAe,EAAE,CAAA;QACjB,OAAM;IACP,CAAC;IACD,GAAG,CAAC,SAAS,CAAC,IAAI,EAAE,8BAA8B,MAAM,CAAC,QAAQ,EAAE,EAAE;QACpE,KAAK,EAAE,MAAM,CAAC,EAAE;KAChB,CAAC,CAAA;IACF,IAAI,SAAuC,CAAA;IAC3C,IAAI,SAA6B,CAAA;IACjC,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,aAAa,CAAC,CAAA;IAC7E,IAAI,CAAC;QACJ,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,CAAC;YACnC,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,sBAAsB,CAAC,CAAA;YAC/C,SAAS,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAuB,CAAA;YACpD,EAAE,CAAC,aAAa,CAAC,aAAa,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAA;QACpE,CAAC;aAAM,CAAC;YACP,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC,CAAA;QAChE,CAAC;QACD,MAAM,cAAc,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE;YAC9C,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;gBAC/B,OAAO,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,GAAG,CAAA;YAClC,CAAC;YACD,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;gBAC5B,OAAO,IAAI,CAAC,GAAG,KAAK,MAAM,CAAC,GAAG,CAAA;YAC/B,CAAC;YACD,OAAO,IAAI,CAAC,QAAQ,KAAK,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAA;QACxD,CAAC,CAAC,CAAA;QACF,IAAI,cAAc,EAAE,CAAC;YACpB,SAAS,GAAG,cAAc,CAAA;QAC3B,CAAC;QAED,MAAM,eAAe,GAAG,MAAM,iBAAiB,CAAC;YAC/C,GAAG,MAAM;YACT,GAAG,EAAE,MAAM,CAAC,GAAG,IAAI,SAAS,EAAE,MAAM;SACpC,CAAC,CAAA;QACF,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,OAAM;QACP,CAAC;QACD,IAAI,eAAe,EAAE,CAAC;YACrB,IAAI,MAAM,CAAC,GAAG,EAAE,CAAC;gBAChB,GAAG,CAAC,SAAS,CAAC,IAAI,EAAE,aAAa,EAAE;oBAClC,KAAK,EAAE,MAAM,CAAC,EAAE;oBAChB,GAAG,EAAE,MAAM,CAAC,GAAG;oBACf,QAAQ,EAAE,MAAM,CAAC,QAAQ;oBACzB,KAAK,EAAE,eAAe;iBACtB,CAAC,CAAA;gBACF,MAAM,iBAAiB,CAAC;oBACvB,GAAG,EAAE,MAAM,CAAC,GAAG;oBACf,QAAQ,EAAE,MAAM,CAAC,QAAQ;oBACzB,eAAe,EAAE,MAAM,CAAC,eAAe;oBACvC,UAAU;oBACV,MAAM;iBACN,CAAC,CAAA;YACH,CAAC;iBAAM,CAAC;gBACP,MAAM,IAAI,KAAK,CAAC,GAAG,eAAe,oBAAoB,CAAC,CAAA;YACxD,CAAC;QACF,CAAC;QAED,MAAM,oBAAoB,GAAG,MAAM,iBAAiB,CAAC;YACpD,GAAG,MAAM;YACT,GAAG,EAAE,MAAM,CAAC,GAAG,IAAI,SAAS,EAAE,MAAM;SACpC,CAAC,CAAA;QACF,IAAI,oBAAoB,EAAE,CAAC;YAC1B,MAAM,IAAI,KAAK,CAAC,iCAAiC,oBAAoB,EAAE,CAAC,CAAA;QACzE,CAAC;QACD,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,OAAM;QACP,CAAC;QAED,OAAO,SAAS,CAAA;IACjB,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,MAAM,KAAK,CAAA;IACZ,CAAC;YAAS,CAAC;QACV,eAAe,EAAE,CAAA;IAClB,CAAC;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CAAC,EAAE,MAAM,EAAE,GAAG,EAAqC,EAAE,MAAoB;IAC5G,GAAG,CAAC,SAAS,CAAC,IAAI,EAAE,sBAAsB,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAA;IAC5D,IAAI,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,IAAI,KAAK,CAAA;IACxC,IAAI,OAAO,MAAM,KAAK,SAAS,EAAE,CAAC;QACjC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAA;IAChC,CAAC;SAAM,IAAI,MAAM,KAAK,MAAM,EAAE,CAAC;QAC9B,MAAM,GAAG,KAAK,CAAA;IACf,CAAC;IACD,MAAM,QAAQ,GAAqB;QAClC,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC;QACxC,qBAAqB;QACrB,eAAe,EAAE,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,cAAc;QAC/D,aAAa,EAAE,KAAK;QACpB,MAAM,EAAE,MAAM;QACd,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,IAAI,GAAG;QACpC,IAAI,EAAE,MAAM,CAAC,WAAW,IAAI,IAAI;QAChC,iBAAiB;QACjB,yCAAyC;KACzC,CAAA;IAED,IAAI,SAAoC,CAAA;IACxC,IAAI,MAAM,CAAC,IAAI,KAAK,iBAAiB,EAAE,CAAC;QACvC,SAAS,GAAG,WAAW,CAAA;IACxB,CAAC;SAAM,IAAI,MAAM,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;QACxC,SAAS,GAAG,WAAW,CAAA;IACxB,CAAC;SAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,0BAA0B,MAAM,CAAC,IAAI,EAAE,CAAC,CAAA;IACzD,CAAC;IAED,MAAM,QAAQ,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE;QAChE,GAAG,QAAQ;QACX,IAAI,EAAE,SAAS;KACf,CAAC,CAAA;IACF,IAAI,MAAM,CAAC,MAAM,EAAE,UAAU,EAAE,CAAC;QAC/B,QAAQ,CAAC,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAA;IACtD,CAAC;IAED,IAAI,UAAU,IAAI,QAAQ,EAAE,CAAC;QAC5B,IAAI,MAAM,CAAC,eAAe,EAAE,MAAM,EAAE,CAAC;YACpC,IAAI,QAAQ,GAAG,sBAAsB,CAAC,MAAM,CAAC,eAAe,CAAC,CAAA;YAC7D,IAAI,YAAY,CAAA;YAChB,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBACnC,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAA;gBAClC,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;YAC7B,CAAC;YACD,MAAM,QAAQ,CAAC,iBAAiB,CAAC;gBAChC,YAAY;gBACZ,QAAQ;aACR,CAAC,CAAA;QACH,CAAC;aAAM,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAC1B,MAAM,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE;gBACtC,QAAQ,EAAE,CAAC;aACX,CAAC,CAAA;QACH,CAAC;aAAM,CAAC;YACP,MAAM,QAAQ,CAAC,QAAQ,CAAC,EAAE,EAAE;gBAC3B,QAAQ,EAAE,CAAC;aACX,CAAC,CAAA;QACH,CAAC;IACF,CAAC;IAED,OAAO,QAAQ,CAAA;AAChB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,QAAyB;IAC9D,QAAQ,CAAC,OAAO,EAAE,CAAA;AACnB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC9C,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAgD,EAC1E,QAAyB,EACzB,MAAoB;IAEpB,IAAI,CAAC,CAAC,UAAU,IAAI,QAAQ,CAAC,EAAE,CAAC;QAC/B,MAAM,IAAI,KAAK,CAAC,4CAA4C,CAAC,CAAA;IAC9D,CAAC;IACD,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;QACrB,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAA;IAC3D,CAAC;IAED,IAAI,YAAY,GAA2B,UAAU,CAAA;IACrD,IAAI,cAAkC,CAAA;IAEtC,MAAM,QAAQ,GAAG,MAAM,CAAC,kBAAkB,IAAI,EAAE,CAAA;IAChD,MAAM,YAAY,GAAG,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,IAAI,EAAE,CAAA;IACxD,MAAM,oBAAoB,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAA;IACzF,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,EAAE;QACtD,aAAa;QACb,OAAO,EAAE,IAAI,EAAE,0DAA0D;QACzE,cAAc,EAAE,IAAI;QACpB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,QAAQ,CAAC,WAAW;QACxD,QAAQ,EAAE,OAAO,CAAC,SAAS,IAAI,QAAQ,CAAC,SAAS;QACjD,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,MAAM,EAAE,MAAM,EAAE,SAAS;QACzB,WAAW,EAAE,OAAO,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAClE,uFAAuF;QACvF,oCAAoC;QACpC,aAAa,EAAE,OAAO,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QAC9D,sEAAsE;QACtE,aAAa;QACb,eAAe,EAAE,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE;YAClC,MAAM,eAAe,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAA;YAClD,IAAI,eAAe,EAAE,CAAC;gBACrB,YAAY,GAAG,aAAa,CAAA;gBAC5B,cAAc,GAAG,IAAI,CAAA;gBACrB,OAAO,KAAK,CAAA;YACb,CAAC;YACD,IAAI,OAAO,EAAE,CAAC;gBACb,OAAO,CAAC;oBACP,IAAI;oBACJ,MAAM,EAAE,CAAC,OAAO,CAAC;iBACjB,CAAC,CAAA;YACH,CAAC;YACD,OAAO,CAAC,MAAM,EAAE,OAAO,CAAA;QACxB,CAAC;QACD,aAAa;QACb,gBAAgB,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE;YACxC,MAAM,eAAe,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAA;YAClD,IAAI,eAAe,EAAE,CAAC;gBACrB,YAAY,GAAG,aAAa,CAAA;gBAC5B,cAAc,GAAG,IAAI,CAAA;gBACrB,OAAO,KAAK,CAAA;YACb,CAAC;YACD,IAAI,OAAO,EAAE,CAAC;gBACb,OAAO,CAAC;oBACP,IAAI;oBACJ,MAAM,EAAE,QAAQ;iBAChB,CAAC,CAAA;YACH,CAAC;YACD,OAAO,CAAC,MAAM,EAAE,OAAO,CAAA;QACxB,CAAC;KACD,CAAC,CAAA;IAEF,IAAI,MAAM,CAAC,eAAe,KAAK,OAAO,CAAC,SAAS,EAAE,CAAC;QAClD,YAAY,GAAG,WAAW,CAAA;IAC3B,CAAC;IAED,IAAI,YAAY,GAAG,MAAM,CAAC,IAAI,CAAA;IAC9B,IAAI,cAAc,EAAE,CAAC;QACpB,YAAY,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,cAAc,CAAC,MAAM,CAAC,CAAA;IAC7D,CAAC;IAED,OAAO;QACN,YAAY;QACZ,IAAI,EAAE,YAAY;QAClB,YAAY,EAAE,MAAM,CAAC,cAAc;QACnC,gBAAgB,EAAE,MAAM,CAAC,eAAe;QACxC,aAAa,EAAE,QAAQ,CAAC,iBAAiB,EAAE,aAAa,CAAC,KAAK,IAAI,CAAC;KACnE,CAAA;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC9C,EAAE,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,GAAG,EAAE,OAAO,EAAgD,EAC7F,QAAyB,EACzB,MAAoB;IAEpB,IAAI,CAAC,CAAC,mBAAmB,IAAI,QAAQ,CAAC,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,4CAA4C,CAAC,CAAA;IAC9D,CAAC;IACD,IAAI,OAAO,GAAG,QAAQ,CAAC,iBAAiB,CAAA;IACxC,IAAI,CAAC,OAAO,IAAI,YAAY,EAAE,CAAC;QAC9B,GAAG,CAAC,SAAS,CAAC,KAAK,EAAE,wBAAwB,CAAC,CAAA;QAC9C,IAAI,QAAQ,GAAG,sBAAsB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAA;QACvD,IAAI,YAAY,CAAA;QAChB,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACnC,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAA;YAClC,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;QAC7B,CAAC;QACD,yBAAyB;QACzB,IAAI,QAAQ,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;YACnD,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAA;QACjC,CAAC;QAED,OAAO,GAAG,MAAM,QAAQ,CAAC,iBAAiB,CAAC;YAC1C,YAAY;YACZ,QAAQ;SACR,CAAC,CAAA;IACH,CAAC;IAED,MAAM,oBAAoB,GAAG,sBAAsB,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAA;IAExG,MAAM,WAAW,GAAG,oBAAoB,CAAC,oBAAoB,CAAC,MAAM,GAAG,CAAC,CAAC,CAAA;IACzE,IAAI,CAAC,CAAC,WAAW,CAAC,IAAI,KAAK,MAAM,IAAI,WAAW,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3D,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAA;IAClE,CAAC;IACD,MAAM,KAAK,GAAoB,WAAW,CAAC,OAAO,CAAA;IAElD,IAAI,YAAY,GAA2B,UAAU,CAAA;IACrD,IAAI,cAAkC,CAAA;IAEtC,MAAM,QAAQ,GAAG,MAAM,CAAC,kBAAkB,IAAI,EAAE,CAAA;IAChD,MAAM,YAAY,GAAG,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,IAAI,EAAE,CAAA;IACxD,MAAM,oBAAoB,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAA;IACzF,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE;QACrD,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,QAAQ,CAAC,WAAW;QACxD,QAAQ,EAAE,OAAO,CAAC,SAAS,IAAI,QAAQ,CAAC,SAAS;QACjD,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI;QACnC,MAAM,EAAE,MAAM,CAAC,SAAS;QACxB,WAAW,EAAE,OAAO,CAAC,gBAAgB,IAAI,QAAQ,CAAC,gBAAgB;QAClE,aAAa,EAAE,OAAO,CAAC,aAAa,IAAI,QAAQ,CAAC,aAAa;QAC9D,0EAA0E;QAC1E,aAAa;QACb,eAAe,EAAE,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE;YAClC,MAAM,eAAe,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAA;YAClD,IAAI,eAAe,EAAE,CAAC;gBACrB,YAAY,GAAG,aAAa,CAAA;gBAC5B,cAAc,GAAG,IAAI,CAAA;gBACrB,OAAO,KAAK,CAAA;YACb,CAAC;YACD,IAAI,OAAO,EAAE,CAAC;gBACb,OAAO,CAAC;oBACP,IAAI;oBACJ,MAAM,EAAE,CAAC,OAAO,CAAC;iBACjB,CAAC,CAAA;YACH,CAAC;YACD,OAAO,CAAC,MAAM,EAAE,OAAO,CAAA;QACxB,CAAC;QACD,aAAa;QACb,gBAAgB,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE;YACxC,MAAM,eAAe,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAA;YAClD,IAAI,eAAe,EAAE,CAAC;gBACrB,YAAY,GAAG,aAAa,CAAA;gBAC5B,cAAc,GAAG,IAAI,CAAA;gBACrB,OAAO,KAAK,CAAA;YACb,CAAC;YACD,IAAI,OAAO,EAAE,CAAC;gBACb,OAAO,CAAC;oBACP,MAAM,EAAE,QAAQ;oBAChB,IAAI;iBACJ,CAAC,CAAA;YACH,CAAC;YAED,OAAO,CAAC,MAAM,EAAE,OAAO,CAAA;QACxB,CAAC;KACD,CAAC,CAAA;IAEF,IAAI,MAAM,CAAC,KAAK,CAAC,iBAAiB,KAAK,OAAO,CAAC,SAAS,EAAE,CAAC;QAC1D,YAAY,GAAG,WAAW,CAAA;IAC3B,CAAC;IAED,IAAI,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAA;IAChD,IAAI,cAAc,EAAE,CAAC;QACpB,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,cAAc,CAAC,MAAM,CAAC,CAAA;IACrD,CAAC;IAED,OAAO;QACN,YAAY;QACZ,OAAO,EAAE;YACR,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,QAAQ;SACjB;QACD,YAAY,EAAE,MAAM,CAAC,KAAK,CAAC,aAAa;QACxC,gBAAgB,EAAE,MAAM,CAAC,KAAK,CAAC,iBAAiB;QAChD,aAAa,EAAE,OAAO,CAAC,aAAa,CAAC,KAAK;KAC1C,CAAA;AACF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACzC,EAAE,OAAO,EAAE,MAAM,EAAuB,EACxC,QAAyB,EACzB,MAAoB;IAEpB,IAAI,CAAC,CAAC,OAAO,IAAI,QAAQ,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAA;IACxD,CAAC;IACD,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAA;IACpD,CAAC;IACD,MAAM,KAAK,GAAa,EAAE,CAAA;IAC1B,IAAI,OAAO,OAAO,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;QACvC,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAA;IAC1B,CAAC;SAAM,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzC,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnC,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;gBAC/B,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;YAClB,CAAC;iBAAM,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAClC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAA;YAC1B,CAAC;iBAAM,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;gBACnC,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAA;YACjD,CAAC;QACF,CAAC;IACF,CAAC;IAED,MAAM,GAAG,GAAG,MAAM,eAAe,CAAC,QAAQ,EAAE,KAAK,EAAE;QAClD,cAAc,EAAE,OAAO,CAAC,UAAU;KAClC,CAAC,CAAA;IAEF,OAAO;QACN,UAAU,EAAE,GAAG,CAAC,UAAU;QAC1B,WAAW,EAAE,GAAG,CAAC,eAAe;KAChC,CAAA;AACF,CAAC"}
@@ -0,0 +1,3 @@
1
+ import { ChatMessage as GPT4AllChatMessage } from 'gpt4all';
2
+ import { ChatMessage } from '../../types/index.js';
3
+ export declare function createChatMessageArray(messages: ChatMessage[]): GPT4AllChatMessage[];
@@ -0,0 +1,29 @@
1
+ import { flattenMessageTextContent } from '../../lib/flattenMessageTextContent.js';
2
+ export function createChatMessageArray(messages) {
3
+ const chatMessages = [];
4
+ let systemPrompt;
5
+ for (const message of messages) {
6
+ if (message.role === 'user' || message.role === 'assistant') {
7
+ chatMessages.push({
8
+ role: message.role,
9
+ content: flattenMessageTextContent(message.content),
10
+ });
11
+ }
12
+ else if (message.role === 'system') {
13
+ if (systemPrompt) {
14
+ systemPrompt += '\n\n' + message.content;
15
+ }
16
+ else {
17
+ systemPrompt = flattenMessageTextContent(message.content);
18
+ }
19
+ }
20
+ }
21
+ if (systemPrompt) {
22
+ chatMessages.unshift({
23
+ role: 'system',
24
+ content: systemPrompt,
25
+ });
26
+ }
27
+ return chatMessages;
28
+ }
29
+ //# sourceMappingURL=util.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"util.js","sourceRoot":"","sources":["../../../src/engines/gpt4all/util.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,yBAAyB,EAAE,MAAM,2CAA2C,CAAA;AAErF,MAAM,UAAU,sBAAsB,CACrC,QAAuB;IAEvB,MAAM,YAAY,GAAyB,EAAE,CAAA;IAC7C,IAAI,YAAgC,CAAA;IACpC,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAChC,IAAI,OAAO,CAAC,IAAI,KAAK,MAAM,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YAC7D,YAAY,CAAC,IAAI,CAAC;gBACjB,IAAI,EAAE,OAAO,CAAC,IAAI;gBAClB,OAAO,EAAE,yBAAyB,CAAC,OAAO,CAAC,OAAO,CAAC;aACnD,CAAC,CAAA;QACH,CAAC;aAAM,IAAI,OAAO,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACtC,IAAI,YAAY,EAAE,CAAC;gBAClB,YAAY,IAAI,MAAM,GAAG,OAAO,CAAC,OAAO,CAAA;YACzC,CAAC;iBAAM,CAAC;gBACP,YAAY,GAAG,yBAAyB,CAAC,OAAO,CAAC,OAAO,CAAC,CAAA;YAC1D,CAAC;QACF,CAAC;IACF,CAAC;IACD,IAAI,YAAY,EAAE,CAAC;QAClB,YAAY,CAAC,OAAO,CAAC;YACpB,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,YAAY;SACrB,CAAC,CAAA;IACH,CAAC;IACD,OAAO,YAAY,CAAA;AACpB,CAAC"}
@@ -0,0 +1,19 @@
1
+ import type { ModelPool } from '../pool.js';
2
+ import type { ModelStore } from '../store.js';
3
+ import { ModelEngine, EngineStartContext } from '../types/index.js';
4
+ export declare const BuiltInEngines: {
5
+ readonly gpt4all: "gpt4all";
6
+ readonly nodeLlamaCpp: "node-llama-cpp";
7
+ readonly transformersJs: "transformers-js";
8
+ readonly stableDiffusionCpp: "stable-diffusion-cpp";
9
+ };
10
+ export type BuiltInEngineName = typeof BuiltInEngines[keyof typeof BuiltInEngines];
11
+ export declare const builtInEngineNames: string[];
12
+ export declare class CustomEngine implements ModelEngine {
13
+ pool: ModelPool;
14
+ store: ModelStore;
15
+ start({ pool, store }: EngineStartContext): Promise<void>;
16
+ prepareModel(): Promise<void>;
17
+ createInstance(): Promise<void>;
18
+ disposeInstance(): Promise<void>;
19
+ }
@@ -0,0 +1,21 @@
1
+ export const BuiltInEngines = {
2
+ gpt4all: 'gpt4all',
3
+ nodeLlamaCpp: 'node-llama-cpp',
4
+ transformersJs: 'transformers-js',
5
+ stableDiffusionCpp: 'stable-diffusion-cpp',
6
+ };
7
+ export const builtInEngineNames = [
8
+ ...Object.values(BuiltInEngines),
9
+ ];
10
+ export class CustomEngine {
11
+ pool;
12
+ store;
13
+ async start({ pool, store }) {
14
+ this.pool = pool;
15
+ this.store = store;
16
+ }
17
+ async prepareModel() { }
18
+ async createInstance() { }
19
+ async disposeInstance() { }
20
+ }
21
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/engines/index.ts"],"names":[],"mappings":"AAIA,MAAM,CAAC,MAAM,cAAc,GAAG;IAC7B,OAAO,EAAE,SAAS;IAClB,YAAY,EAAE,gBAAgB;IAC9B,cAAc,EAAE,iBAAiB;IACjC,kBAAkB,EAAE,sBAAsB;CACjC,CAAA;AAIV,MAAM,CAAC,MAAM,kBAAkB,GAAa;IAC3C,GAAG,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC;CACvB,CAAA;AAEV,MAAM,OAAO,YAAY;IACxB,IAAI,CAAY;IAChB,KAAK,CAAa;IAClB,KAAK,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,KAAK,EAAsB;QAC9C,IAAI,CAAC,IAAI,GAAG,IAAI,CAAA;QAChB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAA;IACnB,CAAC;IACD,KAAK,CAAC,YAAY,KAAI,CAAC;IACvB,KAAK,CAAC,cAAc,KAAI,CAAC;IACzB,KAAK,CAAC,eAAe,KAAI,CAAC;CAC1B"}
@@ -0,0 +1,49 @@
1
+ import { LlamaChat, LlamaModel, LlamaContext, LlamaCompletion, LlamaContextSequence, LlamaGrammar, ChatHistoryItem, LlamaChatResponse, LlamaEmbeddingContext, GgufFileInfo, LLamaChatContextShiftOptions, LlamaContextOptions } from 'node-llama-cpp';
2
+ import { EngineChatCompletionResult, EngineTextCompletionResult, EngineTextCompletionArgs, EngineChatCompletionArgs, EngineContext, ToolDefinition, EngineEmbeddingArgs, EngineEmbeddingResult, FileDownloadProgress, ModelConfig, TextCompletionParams, TextCompletionGrammar, ChatMessage } from '../../types/index.js';
3
+ export interface NodeLlamaCppInstance {
4
+ model: LlamaModel;
5
+ context: LlamaContext;
6
+ chat?: LlamaChat;
7
+ chatHistory: ChatHistoryItem[];
8
+ grammars: Record<string, LlamaGrammar>;
9
+ pendingFunctionCalls: Record<string, any>;
10
+ lastEvaluation?: LlamaChatResponse['lastEvaluation'];
11
+ embeddingContext?: LlamaEmbeddingContext;
12
+ completion?: LlamaCompletion;
13
+ contextSequence: LlamaContextSequence;
14
+ }
15
+ export interface NodeLlamaCppModelMeta {
16
+ gguf: GgufFileInfo;
17
+ }
18
+ export interface NodeLlamaCppModelConfig extends ModelConfig {
19
+ location: string;
20
+ grammars?: Record<string, TextCompletionGrammar>;
21
+ sha256?: string;
22
+ completionDefaults?: TextCompletionParams;
23
+ initialMessages?: ChatMessage[];
24
+ prefix?: string;
25
+ tools?: {
26
+ definitions: Record<string, ToolDefinition>;
27
+ includeToolDocumentation?: boolean;
28
+ parallelism?: number;
29
+ };
30
+ contextSize?: number;
31
+ batchSize?: number;
32
+ lora?: LlamaContextOptions['lora'];
33
+ contextShiftStrategy?: LLamaChatContextShiftOptions['strategy'];
34
+ device?: {
35
+ gpu?: boolean | 'auto' | (string & {});
36
+ gpuLayers?: number;
37
+ cpuThreads?: number;
38
+ memLock?: boolean;
39
+ };
40
+ }
41
+ export declare const autoGpu = true;
42
+ export declare function prepareModel({ config, log }: EngineContext<NodeLlamaCppModelConfig>, onProgress?: (progress: FileDownloadProgress) => void, signal?: AbortSignal): Promise<{
43
+ gguf: GgufFileInfo;
44
+ } | undefined>;
45
+ export declare function createInstance({ config, log }: EngineContext<NodeLlamaCppModelConfig>, signal?: AbortSignal): Promise<NodeLlamaCppInstance>;
46
+ export declare function disposeInstance(instance: NodeLlamaCppInstance): Promise<void>;
47
+ export declare function processChatCompletionTask({ request, config, resetContext, log, onChunk }: EngineChatCompletionArgs<NodeLlamaCppModelConfig>, instance: NodeLlamaCppInstance, signal?: AbortSignal): Promise<EngineChatCompletionResult>;
48
+ export declare function processTextCompletionTask({ request, config, resetContext, log, onChunk }: EngineTextCompletionArgs<NodeLlamaCppModelConfig>, instance: NodeLlamaCppInstance, signal?: AbortSignal): Promise<EngineTextCompletionResult>;
49
+ export declare function processEmbeddingTask({ request, config, log }: EngineEmbeddingArgs<NodeLlamaCppModelConfig>, instance: NodeLlamaCppInstance, signal?: AbortSignal): Promise<EngineEmbeddingResult>;