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 +1,133 @@
1
+ import fs from 'node:fs';
2
+ import { AutoModel, AutoProcessor, AutoTokenizer } from '@huggingface/transformers';
3
+ import { resolveModelFileLocation } from '../../lib/resolveModelFileLocation.js';
4
+ import { parseHuggingfaceModelIdAndBranch, remoteFileExists } from './util.js';
5
+ async function validateModel(modelOpts, config, modelPath) {
6
+ const modelClass = modelOpts.modelClass ?? AutoModel;
7
+ const device = config.device?.gpu ? 'gpu' : 'cpu';
8
+ try {
9
+ const model = await modelClass.from_pretrained(modelPath, {
10
+ local_files_only: true,
11
+ device: device,
12
+ dtype: modelOpts.dtype || 'fp32',
13
+ });
14
+ await model.dispose();
15
+ }
16
+ catch (error) {
17
+ return `Failed to load model (${error})`;
18
+ }
19
+ return undefined;
20
+ }
21
+ async function validateTokenizer(modelOpts, config, modelPath) {
22
+ const tokenizerClass = modelOpts.tokenizerClass ?? AutoTokenizer;
23
+ try {
24
+ await tokenizerClass.from_pretrained(modelPath, {
25
+ local_files_only: true,
26
+ });
27
+ }
28
+ catch (error) {
29
+ return `Failed to load tokenizer (${error})`;
30
+ }
31
+ return undefined;
32
+ }
33
+ async function validateProcessor(modelOpts, config, modelPath) {
34
+ const processorClass = modelOpts.processorClass ?? AutoProcessor;
35
+ try {
36
+ if (modelOpts.processor) {
37
+ const processorPath = resolveModelFileLocation({
38
+ url: modelOpts.processor.url,
39
+ filePath: modelOpts.processor.file,
40
+ modelsCachePath: config.modelsCachePath,
41
+ });
42
+ await processorClass.from_pretrained(processorPath, {
43
+ local_files_only: true,
44
+ });
45
+ }
46
+ else {
47
+ if (modelOpts.processorClass) {
48
+ await processorClass.from_pretrained(modelPath, {
49
+ local_files_only: true,
50
+ });
51
+ }
52
+ else if (config.url) {
53
+ const { branch } = parseHuggingfaceModelIdAndBranch(config.url);
54
+ const [hasProcessor, hasPreprocessor] = await Promise.all([
55
+ remoteFileExists(`${config.url}/blob/${branch}/processor_config.json`),
56
+ remoteFileExists(`${config.url}/blob/${branch}/preprocessor_config.json`),
57
+ ]);
58
+ if (hasProcessor || hasPreprocessor) {
59
+ await processorClass.from_pretrained(modelPath, {
60
+ local_files_only: true,
61
+ });
62
+ }
63
+ }
64
+ }
65
+ }
66
+ catch (error) {
67
+ return `Failed to load processor (${error})`;
68
+ }
69
+ return undefined;
70
+ }
71
+ export async function validateModelFiles(config) {
72
+ if (!fs.existsSync(config.location)) {
73
+ return {
74
+ message: `model directory does not exist: ${config.location}`,
75
+ };
76
+ }
77
+ let modelPath = config.location;
78
+ if (!modelPath.endsWith('/')) {
79
+ modelPath += '/';
80
+ }
81
+ const validateModelComponents = async (modelOpts) => {
82
+ const componentValidationPromises = [
83
+ validateModel(modelOpts, config, modelPath),
84
+ validateTokenizer(modelOpts, config, modelPath),
85
+ validateProcessor(modelOpts, config, modelPath),
86
+ ];
87
+ // if (modelOpts.processor) {
88
+ // componentValidationPromises.push(validateProcessor(modelOpts, config, modelPath))
89
+ // }
90
+ const [model, tokenizer, processor] = await Promise.all(componentValidationPromises);
91
+ const result = {};
92
+ if (model)
93
+ result.model = model;
94
+ if (tokenizer)
95
+ result.tokenizer = tokenizer;
96
+ if (processor)
97
+ result.processor = processor;
98
+ return result;
99
+ };
100
+ const modelValidationPromises = {};
101
+ const noModelConfigured = !config.textModel && !config.visionModel && !config.speechModel;
102
+ if (config.textModel || noModelConfigured) {
103
+ modelValidationPromises.textModel = validateModelComponents(config.textModel || {});
104
+ }
105
+ if (config.visionModel) {
106
+ modelValidationPromises.visionModel = validateModelComponents(config.visionModel);
107
+ }
108
+ if (config.speechModel) {
109
+ modelValidationPromises.speechModel = validateModelComponents(config.speechModel);
110
+ }
111
+ await Promise.all(Object.values(modelValidationPromises));
112
+ const validationErrors = {};
113
+ const textModelErrors = await modelValidationPromises.textModel;
114
+ if (textModelErrors && Object.keys(textModelErrors).length) {
115
+ validationErrors.textModel = textModelErrors;
116
+ }
117
+ const visionModelErrors = await modelValidationPromises.visionModel;
118
+ if (visionModelErrors && Object.keys(visionModelErrors).length) {
119
+ validationErrors.visionModel = visionModelErrors;
120
+ }
121
+ const speechModelErrors = await modelValidationPromises.speechModel;
122
+ if (speechModelErrors && Object.keys(speechModelErrors).length) {
123
+ validationErrors.speechModel = speechModelErrors;
124
+ }
125
+ if (Object.keys(validationErrors).length > 0) {
126
+ return {
127
+ message: 'Failed to validate model components',
128
+ errors: validationErrors,
129
+ };
130
+ }
131
+ return undefined;
132
+ }
133
+ //# sourceMappingURL=validateModelFiles.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"validateModelFiles.js","sourceRoot":"","sources":["../../../src/engines/transformers-js/validateModelFiles.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,SAAS,CAAA;AACxB,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAA;AAEnF,OAAO,EAAE,wBAAwB,EAAE,MAAM,0CAA0C,CAAA;AAEnF,OAAO,EAAE,gCAAgC,EAAE,gBAAgB,EAAE,MAAM,WAAW,CAAA;AAE9E,KAAK,UAAU,aAAa,CAC3B,SAA8B,EAC9B,MAAiC,EACjC,SAAiB;IAEjB,MAAM,UAAU,GAAG,SAAS,CAAC,UAAU,IAAI,SAAS,CAAA;IACpD,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAA;IACjD,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,MAAM,UAAU,CAAC,eAAe,CAAC,SAAS,EAAE;YACzD,gBAAgB,EAAE,IAAI;YACtB,MAAM,EAAE,MAAM;YACd,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,MAAM;SAChC,CAAC,CAAA;QACF,MAAM,KAAK,CAAC,OAAO,EAAE,CAAA;IACtB,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,OAAO,yBAAyB,KAAK,GAAG,CAAA;IACzC,CAAC;IACD,OAAO,SAAS,CAAA;AACjB,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC/B,SAA8B,EAC9B,MAAiC,EACjC,SAAiB;IAEjB,MAAM,cAAc,GAAG,SAAS,CAAC,cAAc,IAAI,aAAa,CAAA;IAChE,IAAI,CAAC;QACJ,MAAM,cAAc,CAAC,eAAe,CAAC,SAAS,EAAE;YAC/C,gBAAgB,EAAE,IAAI;SACtB,CAAC,CAAA;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,OAAO,6BAA6B,KAAK,GAAG,CAAA;IAC7C,CAAC;IACD,OAAO,SAAS,CAAA;AACjB,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC/B,SAA8B,EAC9B,MAAiC,EACjC,SAAiB;IAEjB,MAAM,cAAc,GAAG,SAAS,CAAC,cAAc,IAAI,aAAa,CAAA;IAChE,IAAI,CAAC;QACJ,IAAI,SAAS,CAAC,SAAS,EAAE,CAAC;YACzB,MAAM,aAAa,GAAG,wBAAwB,CAAC;gBAC9C,GAAG,EAAE,SAAS,CAAC,SAAS,CAAC,GAAG;gBAC5B,QAAQ,EAAE,SAAS,CAAC,SAAS,CAAC,IAAI;gBAClC,eAAe,EAAE,MAAM,CAAC,eAAe;aACvC,CAAC,CAAA;YACF,MAAM,cAAc,CAAC,eAAe,CAAC,aAAa,EAAE;gBACnD,gBAAgB,EAAE,IAAI;aACtB,CAAC,CAAA;QACH,CAAC;aAAM,CAAC;YACP,IAAI,SAAS,CAAC,cAAc,EAAE,CAAC;gBAC9B,MAAM,cAAc,CAAC,eAAe,CAAC,SAAS,EAAE;oBAC/C,gBAAgB,EAAE,IAAI;iBACtB,CAAC,CAAA;YACH,CAAC;iBAAM,IAAI,MAAM,CAAC,GAAG,EAAE,CAAC;gBACvB,MAAM,EAAE,MAAM,EAAE,GAAG,gCAAgC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAA;gBAC/D,MAAM,CAAC,YAAY,EAAE,eAAe,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;oBACzD,gBAAgB,CAAC,GAAG,MAAM,CAAC,GAAG,SAAS,MAAM,wBAAwB,CAAC;oBACtE,gBAAgB,CAAC,GAAG,MAAM,CAAC,GAAG,SAAS,MAAM,2BAA2B,CAAC;iBACzE,CAAC,CAAA;gBACF,IAAI,YAAY,IAAI,eAAe,EAAE,CAAC;oBACrC,MAAM,cAAc,CAAC,eAAe,CAAC,SAAS,EAAE;wBAC/C,gBAAgB,EAAE,IAAI;qBACtB,CAAC,CAAA;gBACH,CAAC;YACF,CAAC;QACF,CAAC;IACF,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QAChB,OAAO,6BAA6B,KAAK,GAAG,CAAA;IAC7C,CAAC;IACD,OAAO,SAAS,CAAA;AACjB,CAAC;AAmBD,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACvC,MAAiC;IAEjC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QACrC,OAAO;YACN,OAAO,EAAE,mCAAmC,MAAM,CAAC,QAAQ,EAAE;SAC7D,CAAA;IACF,CAAC;IAED,IAAI,SAAS,GAAG,MAAM,CAAC,QAAQ,CAAA;IAC/B,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QAC9B,SAAS,IAAI,GAAG,CAAA;IACjB,CAAC;IAED,MAAM,uBAAuB,GAAG,KAAK,EAAE,SAA8B,EAAE,EAAE;QACxE,MAAM,2BAA2B,GAAG;YACnC,aAAa,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,CAAC;YAC3C,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,CAAC;YAC/C,iBAAiB,CAAC,SAAS,EAAE,MAAM,EAAE,SAAS,CAAC;SAC/C,CAAA;QAGD,6BAA6B;QAC7B,qFAAqF;QACrF,IAAI;QACJ,MAAM,CAAC,KAAK,EAAE,SAAS,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAA;QACpF,MAAM,MAAM,GAA8B,EAAE,CAAA;QAC5C,IAAI,KAAK;YAAE,MAAM,CAAC,KAAK,GAAG,KAAK,CAAA;QAC/B,IAAI,SAAS;YAAE,MAAM,CAAC,SAAS,GAAG,SAAS,CAAA;QAC3C,IAAI,SAAS;YAAE,MAAM,CAAC,SAAS,GAAG,SAAS,CAAA;QAC3C,OAAO,MAAM,CAAA;IACd,CAAC,CAAA;IAED,MAAM,uBAAuB,GAAQ,EAAE,CAAA;IACvC,MAAM,iBAAiB,GAAG,CAAC,MAAM,CAAC,SAAS,IAAI,CAAC,MAAM,CAAC,WAAW,IAAI,CAAC,MAAM,CAAC,WAAW,CAAA;IACzF,IAAI,MAAM,CAAC,SAAS,IAAI,iBAAiB,EAAE,CAAC;QAC3C,uBAAuB,CAAC,SAAS,GAAG,uBAAuB,CAAC,MAAM,CAAC,SAAS,IAAI,EAAE,CAAC,CAAA;IACpF,CAAC;IACD,IAAI,MAAM,CAAC,WAAW,EAAE,CAAC;QACxB,uBAAuB,CAAC,WAAW,GAAG,uBAAuB,CAAC,MAAM,CAAC,WAAW,CAAC,CAAA;IAClF,CAAC;IACD,IAAI,MAAM,CAAC,WAAW,EAAE,CAAC;QACxB,uBAAuB,CAAC,WAAW,GAAG,uBAAuB,CAAC,MAAM,CAAC,WAAW,CAAC,CAAA;IAClF,CAAC;IAED,MAAM,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC,CAAA;IACzD,MAAM,gBAAgB,GAA0B,EAAE,CAAA;IAClD,MAAM,eAAe,GAAG,MAAM,uBAAuB,CAAC,SAAS,CAAA;IAC/D,IAAI,eAAe,IAAI,MAAM,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,MAAM,EAAE,CAAC;QAC5D,gBAAgB,CAAC,SAAS,GAAG,eAAe,CAAA;IAC7C,CAAC;IACD,MAAM,iBAAiB,GAAG,MAAM,uBAAuB,CAAC,WAAW,CAAA;IACnE,IAAI,iBAAiB,IAAI,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,MAAM,EAAE,CAAC;QAChE,gBAAgB,CAAC,WAAW,GAAG,iBAAiB,CAAA;IACjD,CAAC;IACD,MAAM,iBAAiB,GAAG,MAAM,uBAAuB,CAAC,WAAW,CAAA;IACnE,IAAI,iBAAiB,IAAI,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC,MAAM,EAAE,CAAC;QAChE,gBAAgB,CAAC,WAAW,GAAG,iBAAiB,CAAA;IACjD,CAAC;IAED,IAAI,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9C,OAAO;YACN,OAAO,EAAE,qCAAqC;YAC9C,MAAM,EAAE,gBAAgB;SACxB,CAAA;IACF,CAAC;IACD,OAAO,SAAS,CAAA;AACjB,CAAC"}
@@ -0,0 +1,11 @@
1
+ import { EngineChatCompletionArgs, ModelEngine } from '../types/index.js';
2
+ import { CustomEngine } from '../engines/index.js';
3
+ export declare class ChatWithVisionEngine extends CustomEngine implements ModelEngine {
4
+ imageToTextModel: string;
5
+ chatModel: string;
6
+ constructor({ imageToTextModel, chatModel }: {
7
+ imageToTextModel: string;
8
+ chatModel: string;
9
+ });
10
+ processChatCompletionTask(args: EngineChatCompletionArgs): Promise<import("../types/index.js").EngineChatCompletionResult>;
11
+ }
@@ -0,0 +1,91 @@
1
+ import { CustomEngine } from '../engines/index.js';
2
+ // an experimental engine that replaces images with their descriptions before passing them to a chat model
3
+ export class ChatWithVisionEngine extends CustomEngine {
4
+ imageToTextModel;
5
+ chatModel;
6
+ constructor({ imageToTextModel, chatModel }) {
7
+ super();
8
+ this.imageToTextModel = imageToTextModel;
9
+ this.chatModel = chatModel;
10
+ }
11
+ async processChatCompletionTask(args) {
12
+ const imageTextPromises = [];
13
+ const imageToTextModel = await this.pool.requestInstance({
14
+ model: this.imageToTextModel,
15
+ });
16
+ const messagesWithImageDescriptions = [...args.request.messages];
17
+ for (let m = 0; m < messagesWithImageDescriptions.length; m++) {
18
+ const message = messagesWithImageDescriptions[m];
19
+ if (!Array.isArray(message.content)) {
20
+ continue;
21
+ }
22
+ for (let p = 0; p < message.content.length; p++) {
23
+ const contentPart = message.content[p];
24
+ if (contentPart.type !== 'image') {
25
+ continue;
26
+ }
27
+ imageTextPromises.push(new Promise(async (resolve, reject) => {
28
+ // Florence2 prompt doc
29
+ // "task_prompts_without_inputs": {
30
+ // "<OCR>": "What is the text in the image?",
31
+ // "<OCR_WITH_REGION>": "What is the text in the image, with regions?",
32
+ // "<CAPTION>": "What does the image describe?",
33
+ // "<DETAILED_CAPTION>": "Describe in detail what is shown in the image.",
34
+ // "<MORE_DETAILED_CAPTION>": "Describe with a paragraph what is shown in the image.",
35
+ // "<OD>": "Locate the objects with category name in the image.",
36
+ // "<DENSE_REGION_CAPTION>": "Locate the objects in the image, with their descriptions.",
37
+ // "<REGION_PROPOSAL>": "Locate the region proposals in the image."
38
+ // },
39
+ // "task_prompts_with_input": {
40
+ // "<CAPTION_TO_PHRASE_GROUNDING>": "Locate the phrases in the caption: {input}",
41
+ // "<REFERRING_EXPRESSION_SEGMENTATION>": "Locate {input} in the image with mask",
42
+ // "<REGION_TO_SEGMENTATION>": "What is the polygon mask of region {input}",
43
+ // "<OPEN_VOCABULARY_DETECTION>": "Locate {input} in the image.",
44
+ // "<REGION_TO_CATEGORY>": "What is the region {input}?",
45
+ // "<REGION_TO_DESCRIPTION>": "What does the region {input} describe?",
46
+ // "<REGION_TO_OCR>": "What text is in the region {input}?"
47
+ // }
48
+ // const imageData = await fetch(contentPart.image.url).then((res) => res.arrayBuffer())
49
+ const task = imageToTextModel.instance.processImageToTextTask({
50
+ model: this.imageToTextModel,
51
+ // url: contentPart.url,
52
+ image: contentPart.image,
53
+ prompt: 'What does the image describe?',
54
+ });
55
+ const result = await task.result;
56
+ resolve({
57
+ text: result.text,
58
+ messageIndex: m,
59
+ contentPartIndex: p,
60
+ });
61
+ }));
62
+ }
63
+ }
64
+ const imageTextResults = await Promise.all(imageTextPromises);
65
+ imageToTextModel.release();
66
+ console.debug('Image text results', imageTextResults);
67
+ for (const imageTextResult of imageTextResults) {
68
+ const { text, messageIndex, contentPartIndex } = imageTextResult;
69
+ const message = messagesWithImageDescriptions[messageIndex];
70
+ // if ('type' in message.content[contentPartIndex]) {
71
+ // message.content[contentPartIndex].type = 'text'
72
+ // @ts-ignore
73
+ message.content[contentPartIndex] = {
74
+ type: 'text',
75
+ text: `User uploaded image: ${text}`,
76
+ };
77
+ }
78
+ const chatRequest = { ...args.request, messages: messagesWithImageDescriptions, model: this.chatModel };
79
+ const chatModel = await this.pool.requestInstance(chatRequest);
80
+ const task = chatModel.instance.processChatCompletionTask(chatRequest, {
81
+ onChunk: (chunk) => {
82
+ if (args.onChunk)
83
+ args.onChunk(chunk);
84
+ },
85
+ });
86
+ const result = await task.result;
87
+ chatModel.release();
88
+ return result;
89
+ }
90
+ }
91
+ //# sourceMappingURL=ChatWithVision.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"ChatWithVision.js","sourceRoot":"","sources":["../../src/experiments/ChatWithVision.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AAExD,0GAA0G;AAE1G,MAAM,OAAO,oBAAqB,SAAQ,YAAY;IACrD,gBAAgB,CAAQ;IACxB,SAAS,CAAQ;IAEjB,YAAY,EAAE,gBAAgB,EAAE,SAAS,EAAmD;QAC3F,KAAK,EAAE,CAAA;QACP,IAAI,CAAC,gBAAgB,GAAG,gBAAgB,CAAA;QACxC,IAAI,CAAC,SAAS,GAAG,SAAS,CAAA;IAC3B,CAAC;IAED,KAAK,CAAC,yBAAyB,CAC9B,IAA8B;QAG9B,MAAM,iBAAiB,GAAU,EAAE,CAAA;QACnC,MAAM,gBAAgB,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC;YACxD,KAAK,EAAE,IAAI,CAAC,gBAAgB;SAC5B,CAAC,CAAA;QAEF,MAAM,6BAA6B,GAAG,CAAC,GAAI,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAA;QAEjE,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,6BAA6B,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC/D,MAAM,OAAO,GAAG,6BAA6B,CAAC,CAAC,CAAC,CAAA;YAChD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;gBACrC,SAAQ;YACT,CAAC;YACD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACjD,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAA;gBACtC,IAAI,WAAW,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;oBAClC,SAAQ;gBACT,CAAC;gBACD,iBAAiB,CAAC,IAAI,CAAC,IAAI,OAAO,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE;oBAC5D,uBAAuB;oBACvB,mCAAmC;oBACnC,8CAA8C;oBAC9C,wEAAwE;oBACxE,iDAAiD;oBACjD,2EAA2E;oBAC3E,uFAAuF;oBACvF,kEAAkE;oBAClE,0FAA0F;oBAC1F,oEAAoE;oBACpE,KAAK;oBACL,+BAA+B;oBAC/B,kFAAkF;oBAClF,mFAAmF;oBACnF,6EAA6E;oBAC7E,kEAAkE;oBAClE,0DAA0D;oBAC1D,wEAAwE;oBACxE,4DAA4D;oBAC5D,IAAI;oBACJ,wFAAwF;oBACxF,MAAM,IAAI,GAAG,gBAAgB,CAAC,QAAQ,CAAC,sBAAsB,CAAC;wBAC7D,KAAK,EAAE,IAAI,CAAC,gBAAgB;wBAC5B,wBAAwB;wBACxB,KAAK,EAAE,WAAW,CAAC,KAAK;wBACxB,MAAM,EAAE,+BAA+B;qBACvC,CAAC,CAAA;oBACF,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAA;oBAChC,OAAO,CAAC;wBACP,IAAI,EAAE,MAAM,CAAC,IAAI;wBACjB,YAAY,EAAE,CAAC;wBACf,gBAAgB,EAAE,CAAC;qBACnB,CAAC,CAAA;gBACH,CAAC,CAAC,CAAC,CAAA;YACJ,CAAC;QACF,CAAC;QAED,MAAM,gBAAgB,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAA;QAC7D,gBAAgB,CAAC,OAAO,EAAE,CAAA;QAC1B,OAAO,CAAC,KAAK,CAAC,oBAAoB,EAAE,gBAAgB,CAAC,CAAA;QAErD,KAAK,MAAM,eAAe,IAAI,gBAAgB,EAAE,CAAC;YAChD,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,gBAAgB,EAAE,GAAG,eAAe,CAAA;YAChE,MAAM,OAAO,GAAG,6BAA6B,CAAC,YAAY,CAAC,CAAA;YAC3D,qDAAqD;YACrD,kDAAkD;YAClD,aAAa;YACb,OAAO,CAAC,OAAO,CAAC,gBAAgB,CAAC,GAAG;gBACnC,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,wBAAwB,IAAI,EAAE;aACpC,CAAA;QACF,CAAC;QAED,MAAM,WAAW,GAAG,EAAE,GAAG,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,6BAA6B,EAAE,KAAK,EAAE,IAAI,CAAC,SAAS,EAAE,CAAA;QACvG,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,WAAW,CAAC,CAAA;QAC9D,MAAM,IAAI,GAAG,SAAS,CAAC,QAAQ,CAAC,yBAAyB,CAAC,WAAW,EAAE;YACtE,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE;gBAClB,IAAI,IAAI,CAAC,OAAO;oBAAE,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAA;YACtC,CAAC;SACD,CAAC,CAAA;QACF,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAA;QAChC,SAAS,CAAC,OAAO,EAAE,CAAA;QACnB,OAAO,MAAM,CAAA;IACd,CAAC;CACD"}
@@ -0,0 +1,4 @@
1
+ "use strict";
2
+ // https://huggingface.co/Gustavosta/MagicPrompt-Stable-Diffusion
3
+ // https://huggingface.co/QuantFactory/MagicPrompt-Stable-Diffusion-GGUF
4
+ //# sourceMappingURL=StableDiffPromptGenerator.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"StableDiffPromptGenerator.js","sourceRoot":"","sources":["../../src/experiments/StableDiffPromptGenerator.ts"],"names":[],"mappings":";AAAA,iEAAiE;AACjE,wEAAwE"}
@@ -0,0 +1,18 @@
1
+ import { EngineSpeechToTextArgs, ModelEngine, ToolDefinition } from '../types/index.js';
2
+ import { CustomEngine } from '../engines/index.js';
3
+ type EngineArgs = {
4
+ speechToTextModel: string;
5
+ chatModel: string;
6
+ tools: Record<string, ToolDefinition>;
7
+ };
8
+ export declare class VoiceFunctionCallEngine extends CustomEngine implements ModelEngine {
9
+ speechToTextModel: string;
10
+ chatModel: string;
11
+ tools: Record<string, ToolDefinition>;
12
+ constructor({ speechToTextModel, chatModel, tools }: EngineArgs);
13
+ createTranscription(args: EngineSpeechToTextArgs): Promise<string>;
14
+ processSpeechToTextTask(args: EngineSpeechToTextArgs): Promise<{
15
+ text: string;
16
+ }>;
17
+ }
18
+ export {};
@@ -0,0 +1,51 @@
1
+ import { CustomEngine } from '../engines/index.js';
2
+ // an experimental engine that forwards a transcription to a (function calling) chat model
3
+ export class VoiceFunctionCallEngine extends CustomEngine {
4
+ speechToTextModel;
5
+ chatModel;
6
+ tools;
7
+ constructor({ speechToTextModel, chatModel, tools }) {
8
+ super();
9
+ this.speechToTextModel = speechToTextModel;
10
+ this.chatModel = chatModel;
11
+ this.tools = tools;
12
+ }
13
+ async createTranscription(args) {
14
+ const speechToTextModel = await this.pool.requestInstance({
15
+ model: this.speechToTextModel,
16
+ });
17
+ const transcriptionTask = speechToTextModel.instance.processSpeechToTextTask({
18
+ ...args.request,
19
+ model: this.speechToTextModel,
20
+ });
21
+ const transcription = await transcriptionTask.result;
22
+ speechToTextModel.release();
23
+ return transcription.text;
24
+ }
25
+ async processSpeechToTextTask(args) {
26
+ const [transcription, chatModel] = await Promise.all([
27
+ this.createTranscription(args),
28
+ this.pool.requestInstance({
29
+ model: this.chatModel,
30
+ }),
31
+ ]);
32
+ const chatTask = chatModel.instance.processChatCompletionTask({
33
+ model: this.chatModel,
34
+ tools: this.tools,
35
+ messages: [
36
+ {
37
+ role: 'user',
38
+ content: transcription,
39
+ },
40
+ ],
41
+ }, {
42
+ onChunk: args.onChunk,
43
+ });
44
+ const chatResponse = await chatTask.result;
45
+ chatModel.release();
46
+ return {
47
+ text: chatResponse.message.content,
48
+ };
49
+ }
50
+ }
51
+ //# sourceMappingURL=VoiceFunctionCall.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"VoiceFunctionCall.js","sourceRoot":"","sources":["../../src/experiments/VoiceFunctionCall.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAA;AAQxD,0FAA0F;AAE1F,MAAM,OAAO,uBACZ,SAAQ,YAAY;IAGpB,iBAAiB,CAAQ;IACzB,SAAS,CAAQ;IACjB,KAAK,CAAgC;IAErC,YAAY,EAAE,iBAAiB,EAAE,SAAS,EAAE,KAAK,EAAc;QAC9D,KAAK,EAAE,CAAA;QACP,IAAI,CAAC,iBAAiB,GAAG,iBAAiB,CAAA;QAC1C,IAAI,CAAC,SAAS,GAAG,SAAS,CAAA;QAC1B,IAAI,CAAC,KAAK,GAAG,KAAK,CAAA;IACnB,CAAC;IAED,KAAK,CAAC,mBAAmB,CAAC,IAA4B;QACrD,MAAM,iBAAiB,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC;YACzD,KAAK,EAAE,IAAI,CAAC,iBAAiB;SAC7B,CAAC,CAAA;QACF,MAAM,iBAAiB,GAAG,iBAAiB,CAAC,QAAQ,CAAC,uBAAuB,CAC3E;YACC,GAAG,IAAI,CAAC,OAAO;YACf,KAAK,EAAE,IAAI,CAAC,iBAAiB;SAC7B,CACD,CAAA;QACD,MAAM,aAAa,GAAG,MAAM,iBAAiB,CAAC,MAAM,CAAA;QACpD,iBAAiB,CAAC,OAAO,EAAE,CAAA;QAC3B,OAAO,aAAa,CAAC,IAAI,CAAA;IAC1B,CAAC;IAED,KAAK,CAAC,uBAAuB,CAAC,IAA4B;QACzD,MAAM,CAAC,aAAa,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YACpD,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC;YAC9B,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC;gBACzB,KAAK,EAAE,IAAI,CAAC,SAAS;aACrB,CAAC;SACF,CAAC,CAAA;QACF,MAAM,QAAQ,GAAG,SAAS,CAAC,QAAQ,CAAC,yBAAyB,CAAC;YAC7D,KAAK,EAAE,IAAI,CAAC,SAAS;YACrB,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,QAAQ,EAAE;gBACT;oBACC,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE,aAAa;iBACtB;aACD;SACD,EAAE;YACF,OAAO,EAAE,IAAI,CAAC,OAAO;SACrB,CAAC,CAAA;QACF,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAC,MAAM,CAAA;QAC1C,SAAS,CAAC,OAAO,EAAE,CAAA;QACnB,OAAO;YACN,IAAI,EAAE,YAAY,CAAC,OAAO,CAAC,OAAO;SAClC,CAAA;IACF,CAAC;CACD"}
package/dist/http.d.ts ADDED
@@ -0,0 +1,19 @@
1
+ import http from 'node:http';
2
+ import { ListenOptions } from 'node:net';
3
+ import { LogLevel } from './lib/logger.js';
4
+ import { ModelServer, ModelServerOptions } from './server.js';
5
+ export declare function createOpenAIMiddleware(modelServer: ModelServer): import("express-serve-static-core").Router;
6
+ export declare function createExpressMiddleware(modelServer: ModelServer): import("express-serve-static-core").Router;
7
+ export interface ModelHTTPServerOptions extends ModelServerOptions {
8
+ listen?: ListenOptions;
9
+ logLevel?: LogLevel;
10
+ }
11
+ export declare class ModelHTTPServer {
12
+ httpServer: http.Server;
13
+ modelServer: ModelServer;
14
+ listenOptions: ListenOptions;
15
+ constructor(options: ModelHTTPServerOptions);
16
+ start(): Promise<void>;
17
+ stop(): Promise<void>;
18
+ }
19
+ export declare function startHTTPServer(options: ModelHTTPServerOptions): Promise<ModelHTTPServer>;
package/dist/http.js ADDED
@@ -0,0 +1,54 @@
1
+ import http from 'node:http';
2
+ import express from 'express';
3
+ import cors from 'cors';
4
+ import { createOpenAIRequestHandlers } from './api/openai/index.js';
5
+ import { createAPIMiddleware } from './api/v1/index.js';
6
+ import { ModelServer } from './server.js';
7
+ export function createOpenAIMiddleware(modelServer) {
8
+ const router = express.Router();
9
+ const requestHandlers = createOpenAIRequestHandlers(modelServer);
10
+ router.get('/v1/models', requestHandlers.models);
11
+ router.post('/v1/completions', requestHandlers.completions);
12
+ router.post('/v1/chat/completions', requestHandlers.chatCompletions);
13
+ router.post('/v1/embeddings', requestHandlers.embeddings);
14
+ return router;
15
+ }
16
+ export function createExpressMiddleware(modelServer) {
17
+ const router = express.Router();
18
+ router.get('/', (req, res) => {
19
+ res.json(modelServer.getStatus());
20
+ });
21
+ router.use('/openai', createOpenAIMiddleware(modelServer));
22
+ router.use('/llama', createAPIMiddleware(modelServer));
23
+ return router;
24
+ }
25
+ export class ModelHTTPServer {
26
+ httpServer;
27
+ modelServer;
28
+ listenOptions;
29
+ constructor(options) {
30
+ const { listen, ...modelServerOpts } = options;
31
+ this.modelServer = new ModelServer(modelServerOpts);
32
+ this.listenOptions = listen ?? { port: 3000 };
33
+ const app = express();
34
+ app.use(cors(), express.json({ limit: '50mb' }), createExpressMiddleware(this.modelServer));
35
+ app.set('json spaces', 2);
36
+ this.httpServer = http.createServer(app);
37
+ }
38
+ async start() {
39
+ await this.modelServer.start();
40
+ await new Promise((resolve) => {
41
+ this.httpServer.listen(this.listenOptions, resolve);
42
+ });
43
+ }
44
+ async stop() {
45
+ this.httpServer.close();
46
+ await this.modelServer.stop();
47
+ }
48
+ }
49
+ export async function startHTTPServer(options) {
50
+ const server = new ModelHTTPServer(options);
51
+ await server.start();
52
+ return server;
53
+ }
54
+ //# sourceMappingURL=http.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"http.js","sourceRoot":"","sources":["../src/http.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAA;AAE5B,OAAO,OAAO,MAAM,SAAS,CAAA;AAC7B,OAAO,IAAI,MAAM,MAAM,CAAA;AACvB,OAAO,EAAE,2BAA2B,EAAE,MAAM,8BAA8B,CAAA;AAC1E,OAAO,EAAE,mBAAmB,EAAE,MAAM,0BAA0B,CAAA;AAE9D,OAAO,EAAE,WAAW,EAAwC,MAAM,oBAAoB,CAAA;AAEtF,MAAM,UAAU,sBAAsB,CAAC,WAAwB;IAC9D,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAA;IAC/B,MAAM,eAAe,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAA;IAChE,MAAM,CAAC,GAAG,CAAC,YAAY,EAAE,eAAe,CAAC,MAAM,CAAC,CAAA;IAChD,MAAM,CAAC,IAAI,CAAC,iBAAiB,EAAE,eAAe,CAAC,WAAW,CAAC,CAAA;IAC3D,MAAM,CAAC,IAAI,CAAC,sBAAsB,EAAE,eAAe,CAAC,eAAe,CAAC,CAAA;IACpE,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,eAAe,CAAC,UAAU,CAAC,CAAA;IACzD,OAAO,MAAM,CAAA;AACd,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,WAAwB;IAC/D,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAA;IAC/B,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE;QAC5B,GAAG,CAAC,IAAI,CAAC,WAAW,CAAC,SAAS,EAAE,CAAC,CAAA;IAClC,CAAC,CAAC,CAAA;IACF,MAAM,CAAC,GAAG,CAAC,SAAS,EAAE,sBAAsB,CAAC,WAAW,CAAC,CAAC,CAAA;IAC1D,MAAM,CAAC,GAAG,CAAC,QAAQ,EAAE,mBAAmB,CAAC,WAAW,CAAC,CAAC,CAAA;IACtD,OAAO,MAAM,CAAA;AACd,CAAC;AAOD,MAAM,OAAO,eAAe;IAC3B,UAAU,CAAa;IACvB,WAAW,CAAa;IACxB,aAAa,CAAe;IAE5B,YAAY,OAA+B;QAC1C,MAAM,EAAE,MAAM,EAAE,GAAG,eAAe,EAAE,GAAG,OAAO,CAAA;QAC9C,IAAI,CAAC,WAAW,GAAG,IAAI,WAAW,CAAC,eAAe,CAAC,CAAA;QACnD,IAAI,CAAC,aAAa,GAAG,MAAM,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,CAAA;QAC7C,MAAM,GAAG,GAAG,OAAO,EAAE,CAAA;QACrB,GAAG,CAAC,GAAG,CACN,IAAI,EAAE,EACN,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,EAC/B,uBAAuB,CAAC,IAAI,CAAC,WAAW,CAAC,CACzC,CAAA;QAED,GAAG,CAAC,GAAG,CAAC,aAAa,EAAE,CAAC,CAAC,CAAA;QACzB,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAA;IACzC,CAAC;IAED,KAAK,CAAC,KAAK;QACV,MAAM,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE,CAAA;QAC9B,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;YACnC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,aAAa,EAAE,OAAO,CAAC,CAAA;QACpD,CAAC,CAAC,CAAA;IACH,CAAC;IAED,KAAK,CAAC,IAAI;QACT,IAAI,CAAC,UAAU,CAAC,KAAK,EAAE,CAAA;QACvB,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,CAAA;IAC9B,CAAC;CACD;AAED,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,OAA+B;IACpE,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC,OAAO,CAAC,CAAA;IAC3C,MAAM,MAAM,CAAC,KAAK,EAAE,CAAA;IACpB,OAAO,MAAM,CAAA;AACd,CAAC"}
@@ -0,0 +1,7 @@
1
+ export * from './api/openai/index.js';
2
+ export * from './types/index.js';
3
+ export * from './pool.js';
4
+ export * from './instance.js';
5
+ export * from './store.js';
6
+ export * from './server.js';
7
+ export * from './http.js';
package/dist/index.js ADDED
@@ -0,0 +1,8 @@
1
+ export * from './api/openai/index.js';
2
+ export * from './types/index.js';
3
+ export * from './pool.js';
4
+ export * from './instance.js';
5
+ export * from './store.js';
6
+ export * from './server.js';
7
+ export * from './http.js';
8
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,uBAAuB,CAAA;AACrC,cAAc,kBAAkB,CAAA;AAChC,cAAc,WAAW,CAAA;AACzB,cAAc,eAAe,CAAA;AAC7B,cAAc,YAAY,CAAA;AAC1B,cAAc,aAAa,CAAA;AAC3B,cAAc,WAAW,CAAA"}
@@ -0,0 +1,88 @@
1
+ import { TextCompletionRequest, ChatCompletionRequest, ModelEngine, ModelConfig, ModelInstanceRequest, CompletionProcessingOptions, EmbeddingRequest, ImageToTextRequest, ProcessingOptions, SpeechToTextRequest, SpeechToTextProcessingOptions, EngineChatCompletionResult, EngineTextCompletionResult, TextToImageRequest, ImageToImageRequest } from './types/index.js';
2
+ import { Logger } from './lib/logger.js';
3
+ type ModelInstanceStatus = 'idle' | 'busy' | 'error' | 'loading' | 'preparing';
4
+ interface ModelInstanceOptions extends ModelConfig {
5
+ log?: Logger;
6
+ gpu: boolean;
7
+ }
8
+ export declare class ModelInstance<TEngineState = unknown> {
9
+ id: string;
10
+ status: ModelInstanceStatus;
11
+ modelId: string;
12
+ config: ModelConfig;
13
+ fingerprint: string;
14
+ createdAt: Date;
15
+ lastUsed: number;
16
+ gpu: boolean;
17
+ ttl: number;
18
+ log: Logger;
19
+ private engine;
20
+ private contextIdentity?;
21
+ private needsContextReset;
22
+ private engineInstance?;
23
+ private currentRequest?;
24
+ private shutdownController;
25
+ constructor(engine: ModelEngine, { log, gpu, ...options }: ModelInstanceOptions);
26
+ private generateInstanceId;
27
+ private generateTaskId;
28
+ load(signal?: AbortSignal): Promise<void>;
29
+ dispose(): Promise<void>;
30
+ lock(request: ModelInstanceRequest): void;
31
+ unlock(): void;
32
+ resetContext(): void;
33
+ getContextStateIdentity(): string | undefined;
34
+ hasContextState(): boolean;
35
+ matchesContextState(request: ModelInstanceRequest): boolean;
36
+ matchesRequirements(request: ModelInstanceRequest): boolean;
37
+ private createTaskController;
38
+ processChatCompletionTask(request: ChatCompletionRequest, options?: CompletionProcessingOptions): {
39
+ id: string;
40
+ model: string;
41
+ createdAt: Date;
42
+ result: Promise<EngineChatCompletionResult>;
43
+ cancel: () => void;
44
+ };
45
+ processTextCompletionTask(request: TextCompletionRequest, options?: CompletionProcessingOptions): {
46
+ id: string;
47
+ model: string;
48
+ createdAt: Date;
49
+ cancel: () => void;
50
+ result: Promise<EngineTextCompletionResult>;
51
+ };
52
+ processEmbeddingTask(request: EmbeddingRequest, options?: ProcessingOptions): {
53
+ id: string;
54
+ model: string;
55
+ createdAt: Date;
56
+ cancel: () => void;
57
+ result: Promise<import("./types/index.js").EngineEmbeddingResult>;
58
+ };
59
+ processImageToTextTask(request: ImageToTextRequest, options?: ProcessingOptions): {
60
+ id: string;
61
+ model: string;
62
+ createdAt: Date;
63
+ cancel: () => void;
64
+ result: Promise<import("./types/index.js").EngineImageToTextResult>;
65
+ };
66
+ processImageToImageTask(request: ImageToImageRequest, options?: ProcessingOptions): {
67
+ id: string;
68
+ model: string;
69
+ createdAt: Date;
70
+ cancel: () => void;
71
+ result: Promise<import("./types/index.js").EngineImageToImageResult>;
72
+ };
73
+ processSpeechToTextTask(request: SpeechToTextRequest, options?: SpeechToTextProcessingOptions): {
74
+ id: string;
75
+ model: string;
76
+ createdAt: Date;
77
+ cancel: () => void;
78
+ result: Promise<import("./types/index.js").EngineSpeechToTextResult>;
79
+ };
80
+ processTextToImageTask(request: TextToImageRequest, options?: ProcessingOptions): {
81
+ id: string;
82
+ model: string;
83
+ createdAt: Date;
84
+ cancel: () => void;
85
+ result: Promise<import("./types/index.js").EngineTextToImageResult>;
86
+ };
87
+ }
88
+ export {};