@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
@@ -0,0 +1,10 @@
1
+ import { type LlmConfig } from "../../../../schemas/index";
2
+ /**
3
+ * Converts an LlmConfig into the flat string-keyed map the C++ addon expects.
4
+ *
5
+ * JS-only fields excluded from the output (must NOT be forwarded to the addon):
6
+ * - modelType (schema discriminant, meaningless at C++ level)
7
+ * - system_prompt (JS-side history seeding only; C++ removed --system-prompt in 8189)
8
+ */
9
+ export declare function transformLlmConfig(llmConfig: LlmConfig): Record<string, string>;
10
+ //# sourceMappingURL=transform.d.ts.map
@@ -0,0 +1,34 @@
1
+ import { TOOLS_MODE } from "../../../../schemas/index.js";
2
+ /**
3
+ * Converts an LlmConfig into the flat string-keyed map the C++ addon expects.
4
+ *
5
+ * JS-only fields excluded from the output (must NOT be forwarded to the addon):
6
+ * - modelType (schema discriminant, meaningless at C++ level)
7
+ * - system_prompt (JS-side history seeding only; C++ removed --system-prompt in 8189)
8
+ */
9
+ export function transformLlmConfig(llmConfig) {
10
+ const transformed = JSON.parse(JSON.stringify(llmConfig, (key, v) => key === "modelType" || key === "system_prompt"
11
+ ? undefined
12
+ : key === "stop_sequences"
13
+ ? Array.isArray(v)
14
+ ? v.join(", ")
15
+ : v
16
+ : typeof v === "number" || typeof v === "boolean"
17
+ ? String(v)
18
+ : v).replace(/"([a-z][A-Za-z]*)":/g, (_, key) => `"${key.replace(/[A-Z]/g, (l) => `_${l.toLowerCase()}`)}":`));
19
+ if ("stop_sequences" in transformed) {
20
+ transformed["reverse_prompt"] = transformed["stop_sequences"];
21
+ delete transformed["stop_sequences"];
22
+ }
23
+ if ("opencl_cache_dir" in transformed) {
24
+ transformed["openclCacheDir"] = transformed["opencl_cache_dir"];
25
+ delete transformed["opencl_cache_dir"];
26
+ }
27
+ if ("tools_mode" in transformed) {
28
+ if (transformed["tools_mode"] === TOOLS_MODE.dynamic) {
29
+ transformed["tools_compact"] = "true";
30
+ }
31
+ delete transformed["tools_mode"];
32
+ }
33
+ return transformed;
34
+ }
@@ -31,6 +31,12 @@ export declare const embeddingsPlugin: {
31
31
  integrated: "integrated";
32
32
  dedicated: "dedicated";
33
33
  }>]>>;
34
+ splitMode: import("zod").ZodOptional<import("zod").ZodEnum<{
35
+ none: "none";
36
+ layer: "layer";
37
+ row: "row";
38
+ }>>;
39
+ tensorSplit: import("zod").ZodOptional<import("zod").ZodString>;
34
40
  verbosity: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodLiteral<0>, import("zod").ZodLiteral<1>, import("zod").ZodLiteral<2>, import("zod").ZodLiteral<3>]>>;
35
41
  openclCacheDir: import("zod").ZodOptional<import("zod").ZodString>;
36
42
  }, import("zod/v4/core").$strip>;
@@ -40,6 +46,7 @@ export declare const embeddingsPlugin: {
40
46
  modelId: import("zod").ZodString;
41
47
  text: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodArray<import("zod").ZodString>]>;
42
48
  type: import("zod").ZodLiteral<"embed">;
49
+ requestId: import("zod").ZodOptional<import("zod").ZodString>;
43
50
  }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
44
51
  type: import("zod").ZodLiteral<"embed">;
45
52
  success: import("zod").ZodBoolean;
@@ -29,6 +29,12 @@ function transformEmbedConfig(embedConfig) {
29
29
  ? `${embedConfig.mainGpu}`
30
30
  : embedConfig.mainGpu;
31
31
  }
32
+ if (embedConfig.splitMode) {
33
+ config["split-mode"] = embedConfig.splitMode;
34
+ }
35
+ if (embedConfig.tensorSplit) {
36
+ config["tensor-split"] = embedConfig.tensorSplit;
37
+ }
32
38
  if (typeof embedConfig.verbosity === "number") {
33
39
  config.verbosity = `${embedConfig.verbosity}`;
34
40
  }
@@ -65,11 +71,14 @@ export const embeddingsPlugin = definePlugin({
65
71
  requestSchema: embedRequestSchema,
66
72
  responseSchema: embedResponseSchema,
67
73
  streaming: false,
74
+ // Model-wide hard cancel via `addon.cancel()` on the llama.cpp
75
+ // embedding addon. Compute is interrupted when fired.
76
+ cancel: { scope: "model", hard: true },
68
77
  handler: async function (request) {
69
78
  const embedResult = await embed({
70
79
  modelId: request.modelId,
71
80
  text: request.text,
72
- });
81
+ }, request.requestId);
73
82
  return forwardModelExecution({
74
83
  type: "embed",
75
84
  success: true,
@@ -137,6 +137,9 @@ export declare const nmtPlugin: {
137
137
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
138
138
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
139
139
  embeddings: "embeddings";
140
+ diffusion: "diffusion";
141
+ tts: "tts";
142
+ ocr: "ocr";
140
143
  "llamacpp-completion": "llamacpp-completion";
141
144
  "whispercpp-transcription": "whispercpp-transcription";
142
145
  "llamacpp-embedding": "llamacpp-embedding";
@@ -149,9 +152,6 @@ export declare const nmtPlugin: {
149
152
  whisper: "whisper";
150
153
  nmt: "nmt";
151
154
  parakeet: "parakeet";
152
- tts: "tts";
153
- ocr: "ocr";
154
- diffusion: "diffusion";
155
155
  }>, import("zod").ZodLiteral<"vad">]>>;
156
156
  }, import("zod/v4/core").$strip>]>>;
157
157
  dstVocabSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -167,6 +167,9 @@ export declare const nmtPlugin: {
167
167
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
168
168
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
169
169
  embeddings: "embeddings";
170
+ diffusion: "diffusion";
171
+ tts: "tts";
172
+ ocr: "ocr";
170
173
  "llamacpp-completion": "llamacpp-completion";
171
174
  "whispercpp-transcription": "whispercpp-transcription";
172
175
  "llamacpp-embedding": "llamacpp-embedding";
@@ -179,9 +182,6 @@ export declare const nmtPlugin: {
179
182
  whisper: "whisper";
180
183
  nmt: "nmt";
181
184
  parakeet: "parakeet";
182
- tts: "tts";
183
- ocr: "ocr";
184
- diffusion: "diffusion";
185
185
  }>, import("zod").ZodLiteral<"vad">]>>;
186
186
  }, import("zod/v4/core").$strip>]>>;
187
187
  normalize: import("zod").ZodOptional<import("zod").ZodNumber>;
@@ -210,6 +210,9 @@ export declare const nmtPlugin: {
210
210
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
211
211
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
212
212
  embeddings: "embeddings";
213
+ diffusion: "diffusion";
214
+ tts: "tts";
215
+ ocr: "ocr";
213
216
  "llamacpp-completion": "llamacpp-completion";
214
217
  "whispercpp-transcription": "whispercpp-transcription";
215
218
  "llamacpp-embedding": "llamacpp-embedding";
@@ -222,9 +225,6 @@ export declare const nmtPlugin: {
222
225
  whisper: "whisper";
223
226
  nmt: "nmt";
224
227
  parakeet: "parakeet";
225
- tts: "tts";
226
- ocr: "ocr";
227
- diffusion: "diffusion";
228
228
  }>, import("zod").ZodLiteral<"vad">]>>;
229
229
  }, import("zod/v4/core").$strip>]>;
230
230
  srcVocabSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -240,6 +240,9 @@ export declare const nmtPlugin: {
240
240
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
241
241
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
242
242
  embeddings: "embeddings";
243
+ diffusion: "diffusion";
244
+ tts: "tts";
245
+ ocr: "ocr";
243
246
  "llamacpp-completion": "llamacpp-completion";
244
247
  "whispercpp-transcription": "whispercpp-transcription";
245
248
  "llamacpp-embedding": "llamacpp-embedding";
@@ -252,9 +255,6 @@ export declare const nmtPlugin: {
252
255
  whisper: "whisper";
253
256
  nmt: "nmt";
254
257
  parakeet: "parakeet";
255
- tts: "tts";
256
- ocr: "ocr";
257
- diffusion: "diffusion";
258
258
  }>, import("zod").ZodLiteral<"vad">]>>;
259
259
  }, import("zod/v4/core").$strip>]>>;
260
260
  dstVocabSrc: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -270,6 +270,9 @@ export declare const nmtPlugin: {
270
270
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
271
271
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
272
272
  embeddings: "embeddings";
273
+ diffusion: "diffusion";
274
+ tts: "tts";
275
+ ocr: "ocr";
273
276
  "llamacpp-completion": "llamacpp-completion";
274
277
  "whispercpp-transcription": "whispercpp-transcription";
275
278
  "llamacpp-embedding": "llamacpp-embedding";
@@ -282,9 +285,6 @@ export declare const nmtPlugin: {
282
285
  whisper: "whisper";
283
286
  nmt: "nmt";
284
287
  parakeet: "parakeet";
285
- tts: "tts";
286
- ocr: "ocr";
287
- diffusion: "diffusion";
288
288
  }>, import("zod").ZodLiteral<"vad">]>>;
289
289
  }, import("zod/v4/core").$strip>]>>;
290
290
  normalize: import("zod").ZodOptional<import("zod").ZodNumber>;
@@ -371,6 +371,7 @@ export declare const nmtPlugin: {
371
371
  nmt: "nmt";
372
372
  }>;
373
373
  type: import("zod").ZodLiteral<"translate">;
374
+ requestId: import("zod").ZodOptional<import("zod").ZodString>;
374
375
  }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
375
376
  modelId: import("zod").ZodString;
376
377
  text: import("zod").ZodString;
@@ -383,6 +384,7 @@ export declare const nmtPlugin: {
383
384
  to: import("zod").ZodString;
384
385
  context: import("zod").ZodOptional<import("zod").ZodString>;
385
386
  type: import("zod").ZodLiteral<"translate">;
387
+ requestId: import("zod").ZodOptional<import("zod").ZodString>;
386
388
  }, import("zod/v4/core").$strip>]>, import("zod").ZodObject<{
387
389
  type: import("zod").ZodLiteral<"translate">;
388
390
  token: import("zod").ZodString;
@@ -108,8 +108,12 @@ export const nmtPlugin = definePlugin({
108
108
  requestSchema: translateRequestSchema,
109
109
  responseSchema: translateResponseSchema,
110
110
  streaming: true,
111
+ // nmtcpp does not expose a cancel surface today — SDK falls
112
+ // back to soft-cancel (stop yielding, drop result, skip
113
+ // post-processing; the C++ work runs to completion).
114
+ cancel: { scope: "none" },
111
115
  handler: async function* (request) {
112
- const stream = translate(request);
116
+ const stream = translate(request, request.requestId);
113
117
  try {
114
118
  let result = await stream.next();
115
119
  while (!result.done) {
@@ -34,6 +34,9 @@ export declare const ocrPlugin: {
34
34
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
35
35
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
36
36
  embeddings: "embeddings";
37
+ diffusion: "diffusion";
38
+ tts: "tts";
39
+ ocr: "ocr";
37
40
  "llamacpp-completion": "llamacpp-completion";
38
41
  "whispercpp-transcription": "whispercpp-transcription";
39
42
  "llamacpp-embedding": "llamacpp-embedding";
@@ -46,9 +49,6 @@ export declare const ocrPlugin: {
46
49
  whisper: "whisper";
47
50
  nmt: "nmt";
48
51
  parakeet: "parakeet";
49
- tts: "tts";
50
- ocr: "ocr";
51
- diffusion: "diffusion";
52
52
  }>, import("zod").ZodLiteral<"vad">]>>;
53
53
  }, import("zod/v4/core").$strip>]>>;
54
54
  }, import("zod/v4/core").$strip>;
@@ -89,6 +89,9 @@ export const ocrPlugin = definePlugin({
89
89
  requestSchema: ocrStreamRequestSchema,
90
90
  responseSchema: ocrStreamResponseSchema,
91
91
  streaming: true,
92
+ // ONNX OCR does not expose a cancel surface — SDK falls back
93
+ // to soft-cancel.
94
+ cancel: { scope: "none" },
92
95
  handler: async function* (request) {
93
96
  const stream = ocr({
94
97
  modelId: request.modelId,
@@ -24,6 +24,9 @@ export declare const ttsPlugin: {
24
24
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
25
25
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
26
26
  embeddings: "embeddings";
27
+ diffusion: "diffusion";
28
+ tts: "tts";
29
+ ocr: "ocr";
27
30
  "llamacpp-completion": "llamacpp-completion";
28
31
  "whispercpp-transcription": "whispercpp-transcription";
29
32
  "llamacpp-embedding": "llamacpp-embedding";
@@ -36,9 +39,6 @@ export declare const ttsPlugin: {
36
39
  whisper: "whisper";
37
40
  nmt: "nmt";
38
41
  parakeet: "parakeet";
39
- tts: "tts";
40
- ocr: "ocr";
41
- diffusion: "diffusion";
42
42
  }>, import("zod").ZodLiteral<"vad">]>>;
43
43
  }, import("zod/v4/core").$strip>]>;
44
44
  ttsSpeechEncoderSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -54,6 +54,9 @@ export declare const ttsPlugin: {
54
54
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
55
55
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
56
56
  embeddings: "embeddings";
57
+ diffusion: "diffusion";
58
+ tts: "tts";
59
+ ocr: "ocr";
57
60
  "llamacpp-completion": "llamacpp-completion";
58
61
  "whispercpp-transcription": "whispercpp-transcription";
59
62
  "llamacpp-embedding": "llamacpp-embedding";
@@ -66,9 +69,6 @@ export declare const ttsPlugin: {
66
69
  whisper: "whisper";
67
70
  nmt: "nmt";
68
71
  parakeet: "parakeet";
69
- tts: "tts";
70
- ocr: "ocr";
71
- diffusion: "diffusion";
72
72
  }>, import("zod").ZodLiteral<"vad">]>>;
73
73
  }, import("zod/v4/core").$strip>]>;
74
74
  ttsEmbedTokensSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -84,6 +84,9 @@ export declare const ttsPlugin: {
84
84
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
85
85
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
86
86
  embeddings: "embeddings";
87
+ diffusion: "diffusion";
88
+ tts: "tts";
89
+ ocr: "ocr";
87
90
  "llamacpp-completion": "llamacpp-completion";
88
91
  "whispercpp-transcription": "whispercpp-transcription";
89
92
  "llamacpp-embedding": "llamacpp-embedding";
@@ -96,9 +99,6 @@ export declare const ttsPlugin: {
96
99
  whisper: "whisper";
97
100
  nmt: "nmt";
98
101
  parakeet: "parakeet";
99
- tts: "tts";
100
- ocr: "ocr";
101
- diffusion: "diffusion";
102
102
  }>, import("zod").ZodLiteral<"vad">]>>;
103
103
  }, import("zod/v4/core").$strip>]>;
104
104
  ttsConditionalDecoderSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -114,6 +114,9 @@ export declare const ttsPlugin: {
114
114
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
115
115
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
116
116
  embeddings: "embeddings";
117
+ diffusion: "diffusion";
118
+ tts: "tts";
119
+ ocr: "ocr";
117
120
  "llamacpp-completion": "llamacpp-completion";
118
121
  "whispercpp-transcription": "whispercpp-transcription";
119
122
  "llamacpp-embedding": "llamacpp-embedding";
@@ -126,9 +129,6 @@ export declare const ttsPlugin: {
126
129
  whisper: "whisper";
127
130
  nmt: "nmt";
128
131
  parakeet: "parakeet";
129
- tts: "tts";
130
- ocr: "ocr";
131
- diffusion: "diffusion";
132
132
  }>, import("zod").ZodLiteral<"vad">]>>;
133
133
  }, import("zod/v4/core").$strip>]>;
134
134
  ttsLanguageModelSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -144,6 +144,9 @@ export declare const ttsPlugin: {
144
144
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
145
145
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
146
146
  embeddings: "embeddings";
147
+ diffusion: "diffusion";
148
+ tts: "tts";
149
+ ocr: "ocr";
147
150
  "llamacpp-completion": "llamacpp-completion";
148
151
  "whispercpp-transcription": "whispercpp-transcription";
149
152
  "llamacpp-embedding": "llamacpp-embedding";
@@ -156,9 +159,6 @@ export declare const ttsPlugin: {
156
159
  whisper: "whisper";
157
160
  nmt: "nmt";
158
161
  parakeet: "parakeet";
159
- tts: "tts";
160
- ocr: "ocr";
161
- diffusion: "diffusion";
162
162
  }>, import("zod").ZodLiteral<"vad">]>>;
163
163
  }, import("zod/v4/core").$strip>]>;
164
164
  referenceAudioSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -174,6 +174,9 @@ export declare const ttsPlugin: {
174
174
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
175
175
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
176
176
  embeddings: "embeddings";
177
+ diffusion: "diffusion";
178
+ tts: "tts";
179
+ ocr: "ocr";
177
180
  "llamacpp-completion": "llamacpp-completion";
178
181
  "whispercpp-transcription": "whispercpp-transcription";
179
182
  "llamacpp-embedding": "llamacpp-embedding";
@@ -186,9 +189,6 @@ export declare const ttsPlugin: {
186
189
  whisper: "whisper";
187
190
  nmt: "nmt";
188
191
  parakeet: "parakeet";
189
- tts: "tts";
190
- ocr: "ocr";
191
- diffusion: "diffusion";
192
192
  }>, import("zod").ZodLiteral<"vad">]>>;
193
193
  }, import("zod/v4/core").$strip>]>;
194
194
  }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
@@ -215,6 +215,9 @@ export declare const ttsPlugin: {
215
215
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
216
216
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
217
217
  embeddings: "embeddings";
218
+ diffusion: "diffusion";
219
+ tts: "tts";
220
+ ocr: "ocr";
218
221
  "llamacpp-completion": "llamacpp-completion";
219
222
  "whispercpp-transcription": "whispercpp-transcription";
220
223
  "llamacpp-embedding": "llamacpp-embedding";
@@ -227,9 +230,6 @@ export declare const ttsPlugin: {
227
230
  whisper: "whisper";
228
231
  nmt: "nmt";
229
232
  parakeet: "parakeet";
230
- tts: "tts";
231
- ocr: "ocr";
232
- diffusion: "diffusion";
233
233
  }>, import("zod").ZodLiteral<"vad">]>>;
234
234
  }, import("zod/v4/core").$strip>]>;
235
235
  ttsDurationPredictorSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -245,6 +245,9 @@ export declare const ttsPlugin: {
245
245
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
246
246
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
247
247
  embeddings: "embeddings";
248
+ diffusion: "diffusion";
249
+ tts: "tts";
250
+ ocr: "ocr";
248
251
  "llamacpp-completion": "llamacpp-completion";
249
252
  "whispercpp-transcription": "whispercpp-transcription";
250
253
  "llamacpp-embedding": "llamacpp-embedding";
@@ -257,9 +260,6 @@ export declare const ttsPlugin: {
257
260
  whisper: "whisper";
258
261
  nmt: "nmt";
259
262
  parakeet: "parakeet";
260
- tts: "tts";
261
- ocr: "ocr";
262
- diffusion: "diffusion";
263
263
  }>, import("zod").ZodLiteral<"vad">]>>;
264
264
  }, import("zod/v4/core").$strip>]>;
265
265
  ttsVectorEstimatorSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -275,6 +275,9 @@ export declare const ttsPlugin: {
275
275
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
276
276
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
277
277
  embeddings: "embeddings";
278
+ diffusion: "diffusion";
279
+ tts: "tts";
280
+ ocr: "ocr";
278
281
  "llamacpp-completion": "llamacpp-completion";
279
282
  "whispercpp-transcription": "whispercpp-transcription";
280
283
  "llamacpp-embedding": "llamacpp-embedding";
@@ -287,9 +290,6 @@ export declare const ttsPlugin: {
287
290
  whisper: "whisper";
288
291
  nmt: "nmt";
289
292
  parakeet: "parakeet";
290
- tts: "tts";
291
- ocr: "ocr";
292
- diffusion: "diffusion";
293
293
  }>, import("zod").ZodLiteral<"vad">]>>;
294
294
  }, import("zod/v4/core").$strip>]>;
295
295
  ttsVocoderSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -305,6 +305,9 @@ export declare const ttsPlugin: {
305
305
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
306
306
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
307
307
  embeddings: "embeddings";
308
+ diffusion: "diffusion";
309
+ tts: "tts";
310
+ ocr: "ocr";
308
311
  "llamacpp-completion": "llamacpp-completion";
309
312
  "whispercpp-transcription": "whispercpp-transcription";
310
313
  "llamacpp-embedding": "llamacpp-embedding";
@@ -317,9 +320,6 @@ export declare const ttsPlugin: {
317
320
  whisper: "whisper";
318
321
  nmt: "nmt";
319
322
  parakeet: "parakeet";
320
- tts: "tts";
321
- ocr: "ocr";
322
- diffusion: "diffusion";
323
323
  }>, import("zod").ZodLiteral<"vad">]>>;
324
324
  }, import("zod/v4/core").$strip>]>;
325
325
  ttsUnicodeIndexerSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -335,6 +335,9 @@ export declare const ttsPlugin: {
335
335
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
336
336
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
337
337
  embeddings: "embeddings";
338
+ diffusion: "diffusion";
339
+ tts: "tts";
340
+ ocr: "ocr";
338
341
  "llamacpp-completion": "llamacpp-completion";
339
342
  "whispercpp-transcription": "whispercpp-transcription";
340
343
  "llamacpp-embedding": "llamacpp-embedding";
@@ -347,9 +350,6 @@ export declare const ttsPlugin: {
347
350
  whisper: "whisper";
348
351
  nmt: "nmt";
349
352
  parakeet: "parakeet";
350
- tts: "tts";
351
- ocr: "ocr";
352
- diffusion: "diffusion";
353
353
  }>, import("zod").ZodLiteral<"vad">]>>;
354
354
  }, import("zod/v4/core").$strip>]>;
355
355
  ttsTtsConfigSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -365,6 +365,9 @@ export declare const ttsPlugin: {
365
365
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
366
366
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
367
367
  embeddings: "embeddings";
368
+ diffusion: "diffusion";
369
+ tts: "tts";
370
+ ocr: "ocr";
368
371
  "llamacpp-completion": "llamacpp-completion";
369
372
  "whispercpp-transcription": "whispercpp-transcription";
370
373
  "llamacpp-embedding": "llamacpp-embedding";
@@ -377,9 +380,6 @@ export declare const ttsPlugin: {
377
380
  whisper: "whisper";
378
381
  nmt: "nmt";
379
382
  parakeet: "parakeet";
380
- tts: "tts";
381
- ocr: "ocr";
382
- diffusion: "diffusion";
383
383
  }>, import("zod").ZodLiteral<"vad">]>>;
384
384
  }, import("zod/v4/core").$strip>]>;
385
385
  ttsVoiceStyleSrc: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodObject<{
@@ -395,6 +395,9 @@ export declare const ttsPlugin: {
395
395
  sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
396
396
  addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
397
397
  embeddings: "embeddings";
398
+ diffusion: "diffusion";
399
+ tts: "tts";
400
+ ocr: "ocr";
398
401
  "llamacpp-completion": "llamacpp-completion";
399
402
  "whispercpp-transcription": "whispercpp-transcription";
400
403
  "llamacpp-embedding": "llamacpp-embedding";
@@ -407,9 +410,6 @@ export declare const ttsPlugin: {
407
410
  whisper: "whisper";
408
411
  nmt: "nmt";
409
412
  parakeet: "parakeet";
410
- tts: "tts";
411
- ocr: "ocr";
412
- diffusion: "diffusion";
413
413
  }>, import("zod").ZodLiteral<"vad">]>>;
414
414
  }, import("zod/v4/core").$strip>]>;
415
415
  }, import("zod/v4/core").$strip>]>;
@@ -189,6 +189,9 @@ export const ttsPlugin = definePlugin({
189
189
  requestSchema: ttsRequestSchema,
190
190
  responseSchema: ttsResponseSchema,
191
191
  streaming: true,
192
+ // ONNX TTS does not expose a cancel surface — SDK falls back
193
+ // to soft-cancel.
194
+ cancel: { scope: "none" },
192
195
  handler: async function* (request) {
193
196
  const stream = textToSpeech(request);
194
197
  try {
@@ -226,6 +229,9 @@ export const ttsPlugin = definePlugin({
226
229
  responseSchema: textToSpeechStreamResponseSchema,
227
230
  streaming: true,
228
231
  duplex: true,
232
+ // ONNX TTS does not expose a cancel surface — SDK falls back
233
+ // to soft-cancel.
234
+ cancel: { scope: "none" },
229
235
  handler: async function* (request, inputStream) {
230
236
  const stream = textToSpeechStream(request, inputStream);
231
237
  try {