@qvac/sdk 0.10.1 → 0.11.0

package/dist/schemas/llamacpp-config.d.ts

@@ -30,7 +30,18 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
     }>>;
     "cache-type-k": z.ZodOptional<z.ZodString>;
     "cache-type-v": z.ZodOptional<z.ZodString>;
+    "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        integrated: "integrated";
+        dedicated: "dedicated";
+    }>]>>;
+    "split-mode": z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        layer: "layer";
+        row: "row";
+    }>>;
+    "tensor-split": z.ZodOptional<z.ZodString>;
     openclCacheDir: z.ZodOptional<z.ZodString>;
+    reasoning_budget: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>]>>;
     projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
         name: z.ZodOptional<z.ZodString>;
@@ -44,6 +55,9 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
         sha256Checksum: z.ZodOptional<z.ZodString>;
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
+            diffusion: "diffusion";
+            tts: "tts";
+            ocr: "ocr";
             "llamacpp-completion": "llamacpp-completion";
             "whispercpp-transcription": "whispercpp-transcription";
             "llamacpp-embedding": "llamacpp-embedding";
@@ -56,9 +70,6 @@ export declare const llmConfigBaseSchema: z.ZodObject<{
             whisper: "whisper";
             nmt: "nmt";
             parakeet: "parakeet";
-            tts: "tts";
-            ocr: "ocr";
-            diffusion: "diffusion";
         }>, z.ZodLiteral<"vad">]>>;
     }, z.core.$strip>]>>;
 }, z.core.$strip>;
@@ -94,7 +105,18 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
     }>>;
     "cache-type-k": z.ZodOptional<z.ZodString>;
     "cache-type-v": z.ZodOptional<z.ZodString>;
+    "main-gpu": z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
+        integrated: "integrated";
+        dedicated: "dedicated";
+    }>]>>;
+    "split-mode": z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        layer: "layer";
+        row: "row";
+    }>>;
+    "tensor-split": z.ZodOptional<z.ZodString>;
     openclCacheDir: z.ZodOptional<z.ZodString>;
+    reasoning_budget: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>]>>;
     projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
         src: z.ZodString;
         name: z.ZodOptional<z.ZodString>;
@@ -108,6 +130,9 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         sha256Checksum: z.ZodOptional<z.ZodString>;
         addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
             embeddings: "embeddings";
+            diffusion: "diffusion";
+            tts: "tts";
+            ocr: "ocr";
             "llamacpp-completion": "llamacpp-completion";
             "whispercpp-transcription": "whispercpp-transcription";
             "llamacpp-embedding": "llamacpp-embedding";
@@ -120,9 +145,6 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
             whisper: "whisper";
             nmt: "nmt";
             parakeet: "parakeet";
-            tts: "tts";
-            ocr: "ocr";
-            diffusion: "diffusion";
         }>, z.ZodLiteral<"vad">]>>;
     }, z.core.$strip>]>>;
 }, z.core.$strip>, z.ZodTransform<{
@@ -147,7 +169,11 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
     toolsMode?: "static" | "dynamic" | undefined;
     "cache-type-k"?: string | undefined;
     "cache-type-v"?: string | undefined;
+    "main-gpu"?: number | "integrated" | "dedicated" | undefined;
+    "split-mode"?: "none" | "layer" | "row" | undefined;
+    "tensor-split"?: string | undefined;
     openclCacheDir?: string | undefined;
+    reasoning_budget?: 0 | -1 | undefined;
     projectionModelSrc?: string | {
         src: string;
         name?: string | undefined;
@@ -159,7 +185,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        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;
+        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;
     } | undefined;
 }, {
     ctx_size?: number | undefined;
@@ -183,7 +209,11 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
     toolsMode?: "static" | "dynamic" | undefined;
     "cache-type-k"?: string | undefined;
     "cache-type-v"?: string | undefined;
+    "main-gpu"?: number | "integrated" | "dedicated" | undefined;
+    "split-mode"?: "none" | "layer" | "row" | undefined;
+    "tensor-split"?: string | undefined;
     openclCacheDir?: string | undefined;
+    reasoning_budget?: 0 | -1 | undefined;
     projectionModelSrc?: string | {
         src: string;
         name?: string | undefined;
@@ -195,7 +225,7 @@ export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
         engine?: string | undefined;
         expectedSize?: number | undefined;
         sha256Checksum?: string | undefined;
-        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;
+        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;
     } | undefined;
 }>>;
 export type LlmConfig = z.infer<typeof llmConfigSchema>;
@@ -227,6 +257,12 @@ export declare const embedConfigBaseSchema: z.ZodObject<{
         integrated: "integrated";
         dedicated: "dedicated";
     }>]>>;
+    splitMode: z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        layer: "layer";
+        row: "row";
+    }>>;
+    tensorSplit: z.ZodOptional<z.ZodString>;
     verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
     openclCacheDir: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>;
@@ -264,6 +300,12 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
         integrated: "integrated";
         dedicated: "dedicated";
     }>]>>;
+    splitMode: z.ZodOptional<z.ZodEnum<{
+        none: "none";
+        layer: "layer";
+        row: "row";
+    }>>;
+    tensorSplit: z.ZodOptional<z.ZodString>;
     verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
     openclCacheDir: z.ZodOptional<z.ZodString>;
 }, z.core.$strip>, z.ZodTransform<{
@@ -275,6 +317,8 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
     embdNormalize?: number | undefined;
     flashAttention?: "on" | "off" | "auto" | undefined;
     mainGpu?: number | "integrated" | "dedicated" | undefined;
+    splitMode?: "none" | "layer" | "row" | undefined;
+    tensorSplit?: string | undefined;
     verbosity?: 0 | 3 | 2 | 1 | undefined;
     openclCacheDir?: string | undefined;
 }, {
@@ -286,6 +330,8 @@ export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
     embdNormalize?: number | undefined;
     flashAttention?: "on" | "off" | "auto" | undefined;
     mainGpu?: number | "integrated" | "dedicated" | undefined;
+    splitMode?: "none" | "layer" | "row" | undefined;
+    tensorSplit?: string | undefined;
     verbosity?: 0 | 3 | 2 | 1 | undefined;
     openclCacheDir?: string | undefined;
 }>>;

package/dist/schemas/llamacpp-config.js

@@ -30,6 +30,7 @@ export const llmConfigBaseSchema = z.object({
         z.number().int().min(1), // positive integer: fixed token count
     ])
         .optional(),
+    /** JS-side only: seeds conversation history. Never forwarded to the C++ addon. */
     system_prompt: z.string().optional(),
     no_mmap: z.boolean().optional(),
     verbosity: verbositySchema.optional(),
@@ -45,11 +46,20 @@ export const llmConfigBaseSchema = z.object({
         .optional(),
     "cache-type-k": z.string().optional(),
     "cache-type-v": z.string().optional(),
+    "main-gpu": z
+        .union([z.number().int().min(0), z.enum(["integrated", "dedicated"])])
+        .optional(),
+    "split-mode": z.enum(["none", "layer", "row"]).optional(),
+    "tensor-split": z.string().optional(),
     /**
      * Writable directory for OpenCL kernel binary cache. Required on Android
      * for fast GPU startup.
      */
     openclCacheDir: z.string().optional(),
+    /**
+     * Reasoning channel token budget. `-1` = unrestricted, `0` = disabled.
+     */
+    reasoning_budget: z.union([z.literal(-1), z.literal(0)]).optional(),
     projectionModelSrc: modelSrcInputSchema.optional(),
 });
 // Default values - typed as partial of the config
@@ -76,6 +86,8 @@ export const embedConfigBaseSchema = z.object({
     mainGpu: z
         .union([z.number().int().min(0), z.enum(["integrated", "dedicated"])])
         .optional(),
+    splitMode: z.enum(["none", "layer", "row"]).optional(),
+    tensorSplit: z.string().optional(),
     verbosity: verbositySchema.optional(),
     /**
      * Writable directory for OpenCL kernel binary cache. Required on Android