@qvac/sdk 0.10.2 → 0.11.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (200) hide show
  1. package/NOTICE +92 -54
  2. package/dist/client/api/cancel.d.ts +24 -32
  3. package/dist/client/api/cancel.js +58 -31
  4. package/dist/client/api/client-request-id.d.ts +17 -0
  5. package/dist/client/api/client-request-id.js +27 -0
  6. package/dist/client/api/completion-stream.d.ts +1 -1
  7. package/dist/client/api/completion-stream.js +34 -3
  8. package/dist/client/api/diffusion.d.ts +46 -0
  9. package/dist/client/api/diffusion.js +50 -1
  10. package/dist/client/api/download-asset.d.ts +12 -2
  11. package/dist/client/api/download-asset.js +21 -3
  12. package/dist/client/api/embed.d.ts +8 -4
  13. package/dist/client/api/embed.js +14 -1
  14. package/dist/client/api/get-model-info.d.ts +1 -1
  15. package/dist/client/api/index.d.ts +1 -0
  16. package/dist/client/api/index.js +1 -0
  17. package/dist/client/api/load-model.d.ts +12 -4
  18. package/dist/client/api/load-model.js +25 -1
  19. package/dist/client/api/rag.d.ts +9 -3
  20. package/dist/client/api/rag.js +23 -3
  21. package/dist/client/api/transcribe.d.ts +20 -8
  22. package/dist/client/api/transcribe.js +58 -4
  23. package/dist/client/api/unload-model.d.ts +5 -2
  24. package/dist/client/api/unload-model.js +9 -4
  25. package/dist/client/api/upscale.d.ts +45 -0
  26. package/dist/client/api/upscale.js +99 -0
  27. package/dist/client/rpc/expo-rpc-client.js +2 -1
  28. package/dist/client/rpc/node-rpc-client.js +18 -7
  29. package/dist/client/rpc/rpc-client.d.ts +2 -2
  30. package/dist/client/rpc/rpc-client.js +10 -4
  31. package/dist/client/rpc/rpc-error.d.ts +9 -0
  32. package/dist/client/rpc/rpc-error.js +92 -0
  33. package/dist/examples/cancel-by-request-id.d.ts +38 -0
  34. package/dist/examples/cancel-by-request-id.js +103 -0
  35. package/dist/examples/diffusion-esrgan-upscale.d.ts +2 -0
  36. package/dist/examples/diffusion-esrgan-upscale.js +78 -0
  37. package/dist/examples/download-with-cancel.js +10 -10
  38. package/dist/examples/embed-multi-gpu.d.ts +2 -0
  39. package/dist/examples/embed-multi-gpu.js +46 -0
  40. package/dist/examples/llamacpp-multi-gpu.d.ts +2 -0
  41. package/dist/examples/llamacpp-multi-gpu.js +57 -0
  42. package/dist/examples/rag/rag-hyperdb/cancellation.js +20 -17
  43. package/dist/examples/tools/llamacpp-tools-gemma4.d.ts +2 -0
  44. package/dist/examples/tools/llamacpp-tools-gemma4.js +76 -0
  45. package/dist/examples/tools/llamacpp-tools-qwen35.d.ts +2 -0
  46. package/dist/examples/tools/llamacpp-tools-qwen35.js +70 -0
  47. package/dist/examples/transcription/whispercpp-microphone-conversation.d.ts +2 -0
  48. package/dist/examples/transcription/whispercpp-microphone-conversation.js +102 -0
  49. package/dist/examples/tts/supertonic.js +1 -1
  50. package/dist/expo/plugins/withMobileBundle.d.ts +20 -4
  51. package/dist/expo/plugins/withMobileBundle.js +44 -6
  52. package/dist/index.d.ts +5 -2
  53. package/dist/index.js +21 -1
  54. package/dist/models/registry/models.d.ts +186 -318
  55. package/dist/models/registry/models.js +1522 -1667
  56. package/dist/pear/pre.d.ts +3 -0
  57. package/dist/pear/pre.js +3 -0
  58. package/dist/schemas/cancel.d.ts +167 -28
  59. package/dist/schemas/cancel.js +119 -19
  60. package/dist/schemas/common.d.ts +427 -208
  61. package/dist/schemas/common.js +3 -1
  62. package/dist/schemas/completion-event.d.ts +14 -0
  63. package/dist/schemas/completion-event.js +11 -1
  64. package/dist/schemas/completion-stream.d.ts +15 -1
  65. package/dist/schemas/completion-stream.js +25 -7
  66. package/dist/schemas/download-asset.d.ts +15 -9
  67. package/dist/schemas/download-asset.js +17 -6
  68. package/dist/schemas/embed.d.ts +1 -0
  69. package/dist/schemas/embed.js +5 -0
  70. package/dist/schemas/error.d.ts +31 -0
  71. package/dist/schemas/error.js +29 -1
  72. package/dist/schemas/finetune.d.ts +2 -0
  73. package/dist/schemas/finetune.js +5 -0
  74. package/dist/schemas/get-model-info.d.ts +6 -6
  75. package/dist/schemas/llamacpp-config.d.ts +54 -8
  76. package/dist/schemas/llamacpp-config.js +12 -0
  77. package/dist/schemas/load-model.d.ts +1911 -1260
  78. package/dist/schemas/load-model.js +15 -0
  79. package/dist/schemas/model-src-utils.d.ts +14 -14
  80. package/dist/schemas/model-types.d.ts +10 -10
  81. package/dist/schemas/ocr.d.ts +3 -3
  82. package/dist/schemas/plugin.d.ts +40 -1
  83. package/dist/schemas/plugin.js +10 -1
  84. package/dist/schemas/rag.d.ts +9 -0
  85. package/dist/schemas/rag.js +17 -9
  86. package/dist/schemas/registry.d.ts +30 -30
  87. package/dist/schemas/reload-config.d.ts +14 -14
  88. package/dist/schemas/sdcpp-config.d.ts +131 -17
  89. package/dist/schemas/sdcpp-config.js +163 -1
  90. package/dist/schemas/sdk-config.d.ts +649 -271
  91. package/dist/schemas/sdk-errors-client.d.ts +1 -0
  92. package/dist/schemas/sdk-errors-client.js +5 -0
  93. package/dist/schemas/sdk-errors-server.d.ts +5 -0
  94. package/dist/schemas/sdk-errors-server.js +25 -0
  95. package/dist/schemas/text-to-speech.d.ts +78 -78
  96. package/dist/schemas/transcription-config.d.ts +24 -24
  97. package/dist/schemas/transcription.d.ts +56 -2
  98. package/dist/schemas/transcription.js +27 -0
  99. package/dist/schemas/translate.d.ts +2 -0
  100. package/dist/schemas/translate.js +13 -2
  101. package/dist/schemas/translation-config.d.ts +40 -40
  102. package/dist/schemas/unload-model.d.ts +2 -1
  103. package/dist/schemas/unload-model.js +4 -1
  104. package/dist/scripts/run-unit-tests.js +15 -2
  105. package/dist/server/bare/ops/cancel.d.ts +27 -1
  106. package/dist/server/bare/ops/cancel.js +36 -10
  107. package/dist/server/bare/ops/embed.d.ts +3 -3
  108. package/dist/server/bare/ops/embed.js +144 -28
  109. package/dist/server/bare/ops/kv-cache-utils.d.ts +0 -7
  110. package/dist/server/bare/ops/kv-cache-utils.js +8 -47
  111. package/dist/server/bare/ops/transcribe.d.ts +10 -5
  112. package/dist/server/bare/ops/transcribe.js +196 -41
  113. package/dist/server/bare/ops/translate.d.ts +1 -1
  114. package/dist/server/bare/ops/translate.js +197 -88
  115. package/dist/server/bare/plugins/llamacpp-completion/ops/completion-stream.d.ts +12 -3
  116. package/dist/server/bare/plugins/llamacpp-completion/ops/completion-stream.js +170 -172
  117. package/dist/server/bare/plugins/llamacpp-completion/ops/finetune.js +125 -21
  118. package/dist/server/bare/plugins/llamacpp-completion/ops/kv-cache-session.d.ts +154 -0
  119. package/dist/server/bare/plugins/llamacpp-completion/ops/kv-cache-session.js +448 -0
  120. package/dist/server/bare/plugins/llamacpp-completion/ops/kv-cache-state.d.ts +21 -73
  121. package/dist/server/bare/plugins/llamacpp-completion/ops/kv-cache-state.js +18 -101
  122. package/dist/server/bare/plugins/llamacpp-completion/plugin.d.ts +30 -3
  123. package/dist/server/bare/plugins/llamacpp-completion/plugin.js +157 -85
  124. package/dist/server/bare/plugins/llamacpp-completion/transform.d.ts +10 -0
  125. package/dist/server/bare/plugins/llamacpp-completion/transform.js +34 -0
  126. package/dist/server/bare/plugins/llamacpp-embedding/plugin.d.ts +7 -0
  127. package/dist/server/bare/plugins/llamacpp-embedding/plugin.js +10 -1
  128. package/dist/server/bare/plugins/nmtcpp-translation/plugin.d.ts +17 -15
  129. package/dist/server/bare/plugins/nmtcpp-translation/plugin.js +5 -1
  130. package/dist/server/bare/plugins/onnx-ocr/plugin.d.ts +3 -3
  131. package/dist/server/bare/plugins/onnx-ocr/plugin.js +3 -0
  132. package/dist/server/bare/plugins/onnx-tts/plugin.d.ts +39 -39
  133. package/dist/server/bare/plugins/onnx-tts/plugin.js +6 -0
  134. package/dist/server/bare/plugins/parakeet-transcription/plugin.d.ts +29 -21
  135. package/dist/server/bare/plugins/parakeet-transcription/plugin.js +4 -1
  136. package/dist/server/bare/plugins/sdcpp-generation/ops/diffusion.js +8 -0
  137. package/dist/server/bare/plugins/sdcpp-generation/ops/upscale.d.ts +3 -0
  138. package/dist/server/bare/plugins/sdcpp-generation/ops/upscale.js +40 -0
  139. package/dist/server/bare/plugins/sdcpp-generation/plugin.d.ts +91 -16
  140. package/dist/server/bare/plugins/sdcpp-generation/plugin.js +97 -7
  141. package/dist/server/bare/plugins/whispercpp-transcription/plugin.d.ts +22 -3
  142. package/dist/server/bare/plugins/whispercpp-transcription/plugin.js +50 -14
  143. package/dist/server/bare/rag-hyperdb/index.d.ts +1 -1
  144. package/dist/server/bare/rag-hyperdb/index.js +1 -1
  145. package/dist/server/bare/rag-hyperdb/rag-operation-manager.d.ts +26 -4
  146. package/dist/server/bare/rag-hyperdb/rag-operation-manager.js +65 -28
  147. package/dist/server/bare/runtime/disposable-scope.d.ts +36 -0
  148. package/dist/server/bare/runtime/disposable-scope.js +69 -0
  149. package/dist/server/bare/runtime/index.d.ts +8 -0
  150. package/dist/server/bare/runtime/index.js +3 -0
  151. package/dist/server/bare/runtime/request-context.d.ts +49 -0
  152. package/dist/server/bare/runtime/request-context.js +1 -0
  153. package/dist/server/bare/runtime/request-id.d.ts +12 -0
  154. package/dist/server/bare/runtime/request-id.js +14 -0
  155. package/dist/server/bare/runtime/request-registry-singleton.d.ts +4 -0
  156. package/dist/server/bare/runtime/request-registry-singleton.js +27 -0
  157. package/dist/server/bare/runtime/request-registry.d.ts +136 -0
  158. package/dist/server/bare/runtime/request-registry.js +297 -0
  159. package/dist/server/bare/runtime/with-request-context.d.ts +22 -0
  160. package/dist/server/bare/runtime/with-request-context.js +53 -0
  161. package/dist/server/bare/types/addon-responses.d.ts +11 -1
  162. package/dist/server/rpc/handler-registry.js +26 -7
  163. package/dist/server/rpc/handlers/cancel-delegated.js +28 -31
  164. package/dist/server/rpc/handlers/cancelHandler.d.ts +30 -1
  165. package/dist/server/rpc/handlers/cancelHandler.js +54 -22
  166. package/dist/server/rpc/handlers/delete-cache.d.ts +9 -0
  167. package/dist/server/rpc/handlers/delete-cache.js +13 -15
  168. package/dist/server/rpc/handlers/download-asset.js +133 -40
  169. package/dist/server/rpc/handlers/index.d.ts +2 -0
  170. package/dist/server/rpc/handlers/index.js +2 -0
  171. package/dist/server/rpc/handlers/load-model/download-manager.d.ts +60 -2
  172. package/dist/server/rpc/handlers/load-model/download-manager.js +141 -8
  173. package/dist/server/rpc/handlers/load-model/handler.js +213 -98
  174. package/dist/server/rpc/handlers/load-model/http.js +3 -3
  175. package/dist/server/rpc/handlers/load-model/hyperdrive.js +1 -1
  176. package/dist/server/rpc/handlers/load-model/registry.js +1 -1
  177. package/dist/server/rpc/handlers/load-model/resolve-session.d.ts +20 -1
  178. package/dist/server/rpc/handlers/load-model/resolve-session.js +6 -5
  179. package/dist/server/rpc/handlers/load-model/resolve.d.ts +3 -2
  180. package/dist/server/rpc/handlers/load-model/resolve.js +9 -6
  181. package/dist/server/rpc/handlers/load-model/types.d.ts +24 -0
  182. package/dist/server/rpc/handlers/rag.js +133 -15
  183. package/dist/server/rpc/handlers/upscale-stream.d.ts +3 -0
  184. package/dist/server/rpc/handlers/upscale-stream.js +4 -0
  185. package/dist/server/utils/completion-normalizer.js +8 -0
  186. package/dist/server/utils/tools/dialect.js +4 -0
  187. package/dist/server/utils/tools/parser.js +16 -1
  188. package/dist/server/utils/tools/parsers/gemma4native.d.ts +4 -0
  189. package/dist/server/utils/tools/parsers/gemma4native.js +60 -0
  190. package/dist/server/utils/tools/parsers/qwen35.d.ts +4 -0
  191. package/dist/server/utils/tools/parsers/qwen35.js +120 -0
  192. package/dist/utils/aggregate-events.d.ts +9 -0
  193. package/dist/utils/aggregate-events.js +18 -3
  194. package/dist/utils/decorate-promise.d.ts +34 -0
  195. package/dist/utils/decorate-promise.js +35 -0
  196. package/dist/utils/errors-client.d.ts +3 -0
  197. package/dist/utils/errors-client.js +5 -0
  198. package/dist/utils/errors-server.d.ts +85 -0
  199. package/dist/utils/errors-server.js +97 -0
  200. package/package.json +20 -15
@@ -30,7 +30,18 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
30
30
  }>>;
31
31
  "cache-type-k": z.ZodOptional<z.ZodString>;
32
32
  "cache-type-v": z.ZodOptional<z.ZodString>;
33
+ "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
34
+ integrated: "integrated";
35
+ dedicated: "dedicated";
36
+ }>]>>;
37
+ "split-mode": z.ZodOptional<z.ZodEnum<{
38
+ none: "none";
39
+ layer: "layer";
40
+ row: "row";
41
+ }>>;
42
+ "tensor-split": z.ZodOptional<z.ZodString>;
33
43
  openclCacheDir: z.ZodOptional<z.ZodString>;
44
+ reasoning_budget: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>]>>;
34
45
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
35
46
  src: z.ZodString;
36
47
  name: z.ZodOptional<z.ZodString>;
@@ -44,6 +55,9 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
44
55
  sha256Checksum: z.ZodOptional<z.ZodString>;
45
56
  addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
46
57
  embeddings: "embeddings";
58
+ diffusion: "diffusion";
59
+ tts: "tts";
60
+ ocr: "ocr";
47
61
  "llamacpp-completion": "llamacpp-completion";
48
62
  "whispercpp-transcription": "whispercpp-transcription";
49
63
  "llamacpp-embedding": "llamacpp-embedding";
@@ -56,9 +70,6 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
56
70
  whisper: "whisper";
57
71
  nmt: "nmt";
58
72
  parakeet: "parakeet";
59
- tts: "tts";
60
- ocr: "ocr";
61
- diffusion: "diffusion";
62
73
  }>, z.ZodLiteral<"vad">]>>;
63
74
  }, z.core.$strip>]>>;
64
75
  }, z.core.$strip>;
@@ -94,7 +105,18 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
94
105
  }>>;
95
106
  "cache-type-k": z.ZodOptional<z.ZodString>;
96
107
  "cache-type-v": z.ZodOptional<z.ZodString>;
108
+ "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
109
+ integrated: "integrated";
110
+ dedicated: "dedicated";
111
+ }>]>>;
112
+ "split-mode": z.ZodOptional<z.ZodEnum<{
113
+ none: "none";
114
+ layer: "layer";
115
+ row: "row";
116
+ }>>;
117
+ "tensor-split": z.ZodOptional<z.ZodString>;
97
118
  openclCacheDir: z.ZodOptional<z.ZodString>;
119
+ reasoning_budget: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>]>>;
98
120
  projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
99
121
  src: z.ZodString;
100
122
  name: z.ZodOptional<z.ZodString>;
@@ -108,6 +130,9 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
108
130
  sha256Checksum: z.ZodOptional<z.ZodString>;
109
131
  addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
110
132
  embeddings: "embeddings";
133
+ diffusion: "diffusion";
134
+ tts: "tts";
135
+ ocr: "ocr";
111
136
  "llamacpp-completion": "llamacpp-completion";
112
137
  "whispercpp-transcription": "whispercpp-transcription";
113
138
  "llamacpp-embedding": "llamacpp-embedding";
@@ -120,9 +145,6 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
120
145
  whisper: "whisper";
121
146
  nmt: "nmt";
122
147
  parakeet: "parakeet";
123
- tts: "tts";
124
- ocr: "ocr";
125
- diffusion: "diffusion";
126
148
  }>, z.ZodLiteral<"vad">]>>;
127
149
  }, z.core.$strip>]>>;
128
150
  }, z.core.$strip>, z.ZodTransform<{
@@ -147,7 +169,11 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
147
169
  toolsMode?: "static" | "dynamic" | undefined;
148
170
  "cache-type-k"?: string | undefined;
149
171
  "cache-type-v"?: string | undefined;
172
+ "main-gpu"?: number | "integrated" | "dedicated" | undefined;
173
+ "split-mode"?: "none" | "layer" | "row" | undefined;
174
+ "tensor-split"?: string | undefined;
150
175
  openclCacheDir?: string | undefined;
176
+ reasoning_budget?: 0 | -1 | undefined;
151
177
  projectionModelSrc?: string | {
152
178
  src: string;
153
179
  name?: string | undefined;
@@ -159,7 +185,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
159
185
  engine?: string | undefined;
160
186
  expectedSize?: number | undefined;
161
187
  sha256Checksum?: string | undefined;
162
- addon?: "embeddings" | "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "nmt" | "parakeet" | "tts" | "ocr" | "diffusion" | "vad" | undefined;
188
+ addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "nmt" | "parakeet" | "vad" | undefined;
163
189
  } | undefined;
164
190
  }, {
165
191
  ctx_size?: number | undefined;
@@ -183,7 +209,11 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
183
209
  toolsMode?: "static" | "dynamic" | undefined;
184
210
  "cache-type-k"?: string | undefined;
185
211
  "cache-type-v"?: string | undefined;
212
+ "main-gpu"?: number | "integrated" | "dedicated" | undefined;
213
+ "split-mode"?: "none" | "layer" | "row" | undefined;
214
+ "tensor-split"?: string | undefined;
186
215
  openclCacheDir?: string | undefined;
216
+ reasoning_budget?: 0 | -1 | undefined;
187
217
  projectionModelSrc?: string | {
188
218
  src: string;
189
219
  name?: string | undefined;
@@ -195,7 +225,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
195
225
  engine?: string | undefined;
196
226
  expectedSize?: number | undefined;
197
227
  sha256Checksum?: string | undefined;
198
- addon?: "embeddings" | "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "nmt" | "parakeet" | "tts" | "ocr" | "diffusion" | "vad" | undefined;
228
+ addon?: "embeddings" | "diffusion" | "tts" | "ocr" | "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "nmt" | "parakeet" | "vad" | undefined;
199
229
  } | undefined;
200
230
  }>>;
201
231
  export type LlmConfig = z.infer<typeof llmConfigSchema>;
@@ -227,6 +257,12 @@ export declare const embedConfigBaseSchema: z.ZodObject<{
227
257
  integrated: "integrated";
228
258
  dedicated: "dedicated";
229
259
  }>]>>;
260
+ splitMode: z.ZodOptional<z.ZodEnum<{
261
+ none: "none";
262
+ layer: "layer";
263
+ row: "row";
264
+ }>>;
265
+ tensorSplit: z.ZodOptional<z.ZodString>;
230
266
  verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
231
267
  openclCacheDir: z.ZodOptional<z.ZodString>;
232
268
  }, z.core.$strip>;
@@ -264,6 +300,12 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
264
300
  integrated: "integrated";
265
301
  dedicated: "dedicated";
266
302
  }>]>>;
303
+ splitMode: z.ZodOptional<z.ZodEnum<{
304
+ none: "none";
305
+ layer: "layer";
306
+ row: "row";
307
+ }>>;
308
+ tensorSplit: z.ZodOptional<z.ZodString>;
267
309
  verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
268
310
  openclCacheDir: z.ZodOptional<z.ZodString>;
269
311
  }, z.core.$strip>, z.ZodTransform<{
@@ -275,6 +317,8 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
275
317
  embdNormalize?: number | undefined;
276
318
  flashAttention?: "on" | "off" | "auto" | undefined;
277
319
  mainGpu?: number | "integrated" | "dedicated" | undefined;
320
+ splitMode?: "none" | "layer" | "row" | undefined;
321
+ tensorSplit?: string | undefined;
278
322
  verbosity?: 0 | 3 | 2 | 1 | undefined;
279
323
  openclCacheDir?: string | undefined;
280
324
  }, {
@@ -286,6 +330,8 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
286
330
  embdNormalize?: number | undefined;
287
331
  flashAttention?: "on" | "off" | "auto" | undefined;
288
332
  mainGpu?: number | "integrated" | "dedicated" | undefined;
333
+ splitMode?: "none" | "layer" | "row" | undefined;
334
+ tensorSplit?: string | undefined;
289
335
  verbosity?: 0 | 3 | 2 | 1 | undefined;
290
336
  openclCacheDir?: string | undefined;
291
337
  }>>;
@@ -30,6 +30,7 @@ export const llmConfigBaseSchema = z.object({
30
30
  z.number().int().min(1), // positive integer: fixed token count
31
31
  ])
32
32
  .optional(),
33
+ /** JS-side only: seeds conversation history. Never forwarded to the C++ addon. */
33
34
  system_prompt: z.string().optional(),
34
35
  no_mmap: z.boolean().optional(),
35
36
  verbosity: verbositySchema.optional(),
@@ -45,11 +46,20 @@ export const llmConfigBaseSchema = z.object({
45
46
  .optional(),
46
47
  "cache-type-k": z.string().optional(),
47
48
  "cache-type-v": z.string().optional(),
49
+ "main-gpu": z
50
+ .union([z.number().int().min(0), z.enum(["integrated", "dedicated"])])
51
+ .optional(),
52
+ "split-mode": z.enum(["none", "layer", "row"]).optional(),
53
+ "tensor-split": z.string().optional(),
48
54
  /**
49
55
  * Writable directory for OpenCL kernel binary cache. Required on Android
50
56
  * for fast GPU startup.
51
57
  */
52
58
  openclCacheDir: z.string().optional(),
59
+ /**
60
+ * Reasoning channel token budget. `-1` = unrestricted, `0` = disabled.
61
+ */
62
+ reasoning_budget: z.union([z.literal(-1), z.literal(0)]).optional(),
53
63
  projectionModelSrc: modelSrcInputSchema.optional(),
54
64
  });
55
65
  // Default values - typed as partial of the config
@@ -76,6 +86,8 @@ export const embedConfigBaseSchema = z.object({
76
86
  mainGpu: z
77
87
  .union([z.number().int().min(0), z.enum(["integrated", "dedicated"])])
78
88
  .optional(),
89
+ splitMode: z.enum(["none", "layer", "row"]).optional(),
90
+ tensorSplit: z.string().optional(),
79
91
  verbosity: verbositySchema.optional(),
80
92
  /**
81
93
  * Writable directory for OpenCL kernel binary cache. Required on Android