@qvac/sdk 0.8.3 → 0.9.1

package/dist/schemas/common.js

@@ -5,7 +5,7 @@ import { completionStreamRequestSchema, completionStreamResponseSchema, } from "
 import { loadModelRequestSchema, loadModelResponseSchema, modelProgressUpdateSchema, } from "./load-model.js";
 import { downloadAssetRequestSchema, downloadAssetResponseSchema, } from "./download-asset.js";
 import { unloadModelRequestSchema, unloadModelResponseSchema, } from "./unload-model.js";
-import { transcribeStreamRequestSchema, transcribeStreamResponseSchema, } from "./transcription.js";
+import { transcribeRequestSchema, transcribeResponseSchema, transcribeStreamRequestSchema, transcribeStreamResponseSchema, } from "./transcription.js";
 import { embedRequestSchema, embedResponseSchema } from "./embed.js";
 import { cancelRequestSchema, cancelResponseSchema } from "./cancel.js";
 import { provideRequestSchema, provideResponseSchema } from "./provide.js";
@@ -18,14 +18,19 @@ import { ragRequestSchema, ragResponseSchema, ragProgressUpdateSchema, } from ".
 import { deleteCacheRequestSchema, deleteCacheResponseSchema, } from "./delete-cache.js";
 import { getModelInfoRequestSchema, getModelInfoResponseSchema, } from "./get-model-info.js";
 import { ocrStreamRequestSchema, ocrStreamResponseSchema } from "./ocr.js";
+import { diffusionStreamRequestSchema, diffusionStreamResponseSchema, } from "./sdcpp-config.js";
+import { finetuneRequestSchema, finetuneResponseSchema, finetuneProgressResponseSchema, } from "./finetune.js";
 import { pluginInvokeRequestSchema, pluginInvokeResponseSchema, pluginInvokeStreamRequestSchema, pluginInvokeStreamResponseSchema, } from "./plugin.js";
 import { modelRegistryListRequestSchema, modelRegistryListResponseSchema, modelRegistrySearchRequestSchema, modelRegistrySearchResponseSchema, modelRegistryGetModelRequestSchema, modelRegistryGetModelResponseSchema, } from "./registry.js";
+import { suspendRequestSchema, suspendResponseSchema } from "./suspend.js";
+import { resumeRequestSchema, resumeResponseSchema } from "./resume.js";
 export const requestSchema = z.union([
     heartbeatRequestSchema,
     loadModelRequestSchema,
     downloadAssetRequestSchema,
     completionStreamRequestSchema,
     unloadModelRequestSchema,
+    transcribeRequestSchema,
     transcribeStreamRequestSchema,
     loggingStreamRequestSchema,
     embedRequestSchema,
@@ -38,11 +43,15 @@ export const requestSchema = z.union([
     deleteCacheRequestSchema,
     getModelInfoRequestSchema,
     ocrStreamRequestSchema,
+    diffusionStreamRequestSchema,
+    finetuneRequestSchema,
     pluginInvokeRequestSchema,
     pluginInvokeStreamRequestSchema,
     modelRegistryListRequestSchema,
     modelRegistrySearchRequestSchema,
     modelRegistryGetModelRequestSchema,
+    suspendRequestSchema,
+    resumeRequestSchema,
 ]);
 export const responseSchema = z.discriminatedUnion("type", [
     heartbeatResponseSchema,
@@ -51,6 +60,7 @@ export const responseSchema = z.discriminatedUnion("type", [
     completionStreamResponseSchema,
     unloadModelResponseSchema,
     modelProgressUpdateSchema,
+    transcribeResponseSchema,
     transcribeStreamResponseSchema,
     loggingStreamResponseSchema,
     embedResponseSchema,
@@ -65,11 +75,16 @@ export const responseSchema = z.discriminatedUnion("type", [
     deleteCacheResponseSchema,
     getModelInfoResponseSchema,
     ocrStreamResponseSchema,
+    diffusionStreamResponseSchema,
+    finetuneResponseSchema,
+    finetuneProgressResponseSchema,
     pluginInvokeResponseSchema,
     pluginInvokeStreamResponseSchema,
     modelRegistryListResponseSchema,
     modelRegistrySearchResponseSchema,
     modelRegistryGetModelResponseSchema,
+    suspendResponseSchema,
+    resumeResponseSchema,
 ]);
 export const rpcOptionsSchema = z.object({
     timeout: z.number().min(100).optional(),

package/dist/schemas/completion-stream.d.ts

@@ -111,18 +111,26 @@ export declare const completionStreamRequestSchema: z.ZodObject<{
     type: z.ZodLiteral<"completionStream">;
 }, z.core.$strip>;
 export declare const completionStatsSchema: z.ZodObject<{
-    timeToFirstToken: z.ZodNumber;
-    tokensPerSecond: z.ZodNumber;
-    cacheTokens: z.ZodNumber;
+    timeToFirstToken: z.ZodOptional<z.ZodNumber>;
+    tokensPerSecond: z.ZodOptional<z.ZodNumber>;
+    cacheTokens: z.ZodOptional<z.ZodNumber>;
+    backendDevice: z.ZodOptional<z.ZodEnum<{
+        cpu: "cpu";
+        gpu: "gpu";
+    }>>;
 }, z.core.$strip>;
 export declare const completionStreamResponseSchema: z.ZodObject<{
     type: z.ZodLiteral<"completionStream">;
     token: z.ZodString;
     done: z.ZodOptional<z.ZodBoolean>;
     stats: z.ZodOptional<z.ZodObject<{
-        timeToFirstToken: z.ZodNumber;
-        tokensPerSecond: z.ZodNumber;
-        cacheTokens: z.ZodNumber;
+        timeToFirstToken: z.ZodOptional<z.ZodNumber>;
+        tokensPerSecond: z.ZodOptional<z.ZodNumber>;
+        cacheTokens: z.ZodOptional<z.ZodNumber>;
+        backendDevice: z.ZodOptional<z.ZodEnum<{
+            cpu: "cpu";
+            gpu: "gpu";
+        }>>;
     }, z.core.$strip>>;
     toolCallEvent: z.ZodOptional<z.ZodUnion<readonly [z.ZodObject<{
         type: z.ZodLiteral<"toolCall">;

package/dist/schemas/completion-stream.js

@@ -38,9 +38,10 @@ export const completionStreamRequestSchema = completionClientParamsSchema.extend
     type: z.literal("completionStream"),
 });
 export const completionStatsSchema = z.object({
-    timeToFirstToken: z.number(),
-    tokensPerSecond: z.number(),
-    cacheTokens: z.number(),
+    timeToFirstToken: z.number().optional(),
+    tokensPerSecond: z.number().optional(),
+    cacheTokens: z.number().optional(),
+    backendDevice: z.enum(["cpu", "gpu"]).optional(),
 });
 export const completionStreamResponseSchema = z.object({
     type: z.literal("completionStream"),

package/dist/schemas/download-asset.d.ts

@@ -20,6 +20,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
             "onnx-tts": "onnx-tts";
             "parakeet-transcription": "parakeet-transcription";
             "onnx-ocr": "onnx-ocr";
+            "sdcpp-generation": "sdcpp-generation";
             llm: "llm";
             whisper: "whisper";
             embeddings: "embeddings";
@@ -27,6 +28,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
             parakeet: "parakeet";
             tts: "tts";
             ocr: "ocr";
+            diffusion: "diffusion";
         }>, z.ZodLiteral<"vad">]>>;
     }, z.core.$strip>]>;
     seed: z.ZodOptional<z.ZodBoolean>;
@@ -43,7 +45,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "vad" | undefined;
+        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "diffusion" | "vad" | undefined;
     };
 }, {
     assetSrc: string | {
@@ -57,7 +59,7 @@ export declare const downloadAssetOptionsSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "vad" | undefined;
+        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "diffusion" | "vad" | undefined;
     };
     seed?: boolean | undefined;
 }>>;
@@ -81,6 +83,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
             "onnx-tts": "onnx-tts";
             "parakeet-transcription": "parakeet-transcription";
             "onnx-ocr": "onnx-ocr";
+            "sdcpp-generation": "sdcpp-generation";
             llm: "llm";
             whisper: "whisper";
             embeddings: "embeddings";
@@ -88,6 +91,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
             parakeet: "parakeet";
             tts: "tts";
             ocr: "ocr";
+            diffusion: "diffusion";
         }>, z.ZodLiteral<"vad">]>>;
     }, z.core.$strip>]>;
     seed: z.ZodOptional<z.ZodBoolean>;
@@ -110,7 +114,7 @@ export declare const downloadAssetOptionsToRequestSchema: z.ZodPipe<z.ZodObject<
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "vad" | undefined;
+        addon?: "llamacpp-completion" | "whispercpp-transcription" | "llamacpp-embedding" | "nmtcpp-translation" | "onnx-tts" | "parakeet-transcription" | "onnx-ocr" | "sdcpp-generation" | "llm" | "whisper" | "embeddings" | "nmt" | "parakeet" | "tts" | "ocr" | "diffusion" | "vad" | undefined;
     };
     seed?: boolean | undefined;
     onProgress?: unknown;

package/dist/schemas/embed.d.ts

@@ -8,13 +8,32 @@ export declare const embedRequestSchema: z.ZodObject<{
     text: z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodString>]>;
     type: z.ZodLiteral<"embed">;
 }, z.core.$strip>;
+export declare const embedStatsSchema: z.ZodObject<{
+    totalTime: z.ZodOptional<z.ZodNumber>;
+    tokensPerSecond: z.ZodOptional<z.ZodNumber>;
+    totalTokens: z.ZodOptional<z.ZodNumber>;
+    backendDevice: z.ZodOptional<z.ZodEnum<{
+        cpu: "cpu";
+        gpu: "gpu";
+    }>>;
+}, z.core.$strip>;
 export declare const embedResponseSchema: z.ZodObject<{
     type: z.ZodLiteral<"embed">;
     success: z.ZodBoolean;
     embedding: z.ZodDefault<z.ZodUnion<readonly [z.ZodArray<z.ZodNumber>, z.ZodArray<z.ZodArray<z.ZodNumber>>]>>;
+    stats: z.ZodOptional<z.ZodObject<{
+        totalTime: z.ZodOptional<z.ZodNumber>;
+        tokensPerSecond: z.ZodOptional<z.ZodNumber>;
+        totalTokens: z.ZodOptional<z.ZodNumber>;
+        backendDevice: z.ZodOptional<z.ZodEnum<{
+            cpu: "cpu";
+            gpu: "gpu";
+        }>>;
+    }, z.core.$strip>>;
     error: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>;
 export type EmbedParams = z.infer<typeof embedParamsSchema>;
 export type EmbedRequest = z.infer<typeof embedRequestSchema>;
 export type EmbedResponse = z.infer<typeof embedResponseSchema>;
+export type EmbedStats = z.infer<typeof embedStatsSchema>;
 //# sourceMappingURL=embed.d.ts.map

package/dist/schemas/embed.js

@@ -11,11 +11,18 @@ export const embedParamsSchema = z.object({
 export const embedRequestSchema = embedParamsSchema.extend({
     type: z.literal("embed"),
 });
+export const embedStatsSchema = z.object({
+    totalTime: z.number().optional(),
+    tokensPerSecond: z.number().optional(),
+    totalTokens: z.number().optional(),
+    backendDevice: z.enum(["cpu", "gpu"]).optional(),
+});
 export const embedResponseSchema = z.object({
     type: z.literal("embed"),
     success: z.boolean(),
     embedding: z
         .union([z.array(z.number()), z.array(z.array(z.number()))])
         .default([]),
+    stats: embedStatsSchema.optional(),
     error: z.string().optional(),
 });

package/dist/schemas/engine-addon-map.js

@@ -1,5 +1,5 @@
 import { ModelType } from "./model-types.js";
-import { ADDON_EMBEDDING, ADDON_LLM, ADDON_NMT, ADDON_OCR, ADDON_PARAKEET, ADDON_TTS, ADDON_WHISPER, } from "./plugin.js";
+import { ADDON_DIFFUSION, ADDON_EMBEDDING, ADDON_LLM, ADDON_NMT, ADDON_OCR, ADDON_PARAKEET, ADDON_TTS, ADDON_WHISPER, } from "./plugin.js";
 import { modelRegistryEngineSchema, } from "./registry.js";
 // Canonical engine → addon mapping (exhaustive).
 // TypeScript enforces that every ModelRegistryEngine has an entry.
@@ -11,6 +11,7 @@ export const ENGINE_TO_ADDON = {
     [ModelType.onnxTts]: "tts",
     [ModelType.onnxOcr]: "ocr",
     [ModelType.parakeetTranscription]: "parakeet",
+    [ModelType.sdcppGeneration]: "diffusion",
     "onnx-vad": "vad",
 };
 // Legacy engine names → canonical engine.
@@ -34,6 +35,8 @@ const LEGACY_ENGINE_TO_CANONICAL = {
     vad: "onnx-vad",
     tts: ModelType.onnxTts,
     ocr: ModelType.onnxOcr,
+    [ADDON_DIFFUSION]: ModelType.sdcppGeneration,
+    diffusion: ModelType.sdcppGeneration,
 };
 // Resolves any engine string (legacy or canonical) to a validated canonical engine.
 // Returns null if the engine is not recognized.