@qvac/sdk 0.8.0 → 1.0.0

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

@@ -1,263 +1,32 @@
 import { z } from "zod";
 export declare const VERBOSITY: {
-    readonly ERROR: 0;
-    readonly WARN: 1;
-    readonly INFO: 2;
-    readonly DEBUG: 3;
+    readonly SILENT: -1;
+    readonly INFO: 0;
+    readonly DEBUG: 1;
 };
-export declare const llmConfigBaseSchema: z.ZodObject<{
-    ctx_size: z.ZodOptional<z.ZodNumber>;
+export declare const llmConfigSchema: z.ZodObject<{
+    ctx_size: z.ZodDefault<z.ZodNumber>;
     temp: z.ZodOptional<z.ZodNumber>;
     top_p: z.ZodOptional<z.ZodNumber>;
     top_k: z.ZodOptional<z.ZodNumber>;
     seed: z.ZodOptional<z.ZodNumber>;
-    gpu_layers: z.ZodOptional<z.ZodNumber>;
+    gpu_layers: z.ZodDefault<z.ZodNumber>;
     lora: z.ZodOptional<z.ZodString>;
-    device: z.ZodOptional<z.ZodString>;
+    device: z.ZodDefault<z.ZodString>;
     predict: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<-2>, z.ZodNumber]>>;
-    system_prompt: z.ZodOptional<z.ZodString>;
+    system_prompt: z.ZodDefault<z.ZodString>;
     no_mmap: z.ZodOptional<z.ZodBoolean>;
-    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
+    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<0>, z.ZodLiteral<1>]>>;
     presence_penalty: z.ZodOptional<z.ZodNumber>;
     frequency_penalty: z.ZodOptional<z.ZodNumber>;
     repeat_penalty: z.ZodOptional<z.ZodNumber>;
     stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
     n_discarded: z.ZodOptional<z.ZodNumber>;
     tools: z.ZodOptional<z.ZodBoolean>;
-    projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
-        src: z.ZodString;
-        name: z.ZodOptional<z.ZodString>;
-        modelId: z.ZodOptional<z.ZodString>;
-        registryPath: z.ZodOptional<z.ZodString>;
-        registrySource: z.ZodOptional<z.ZodString>;
-        blobCoreKey: z.ZodOptional<z.ZodString>;
-        blobIndex: z.ZodOptional<z.ZodNumber>;
-        engine: z.ZodOptional<z.ZodString>;
-        expectedSize: z.ZodOptional<z.ZodNumber>;
-        sha256Checksum: z.ZodOptional<z.ZodString>;
-        addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
-            "llamacpp-completion": "llamacpp-completion";
-            "whispercpp-transcription": "whispercpp-transcription";
-            "llamacpp-embedding": "llamacpp-embedding";
-            "nmtcpp-translation": "nmtcpp-translation";
-            "onnx-tts": "onnx-tts";
-            "parakeet-transcription": "parakeet-transcription";
-            "onnx-ocr": "onnx-ocr";
-            llm: "llm";
-            whisper: "whisper";
-            embeddings: "embeddings";
-            nmt: "nmt";
-            parakeet: "parakeet";
-            tts: "tts";
-            ocr: "ocr";
-        }>, z.ZodLiteral<"vad">]>>;
-    }, z.core.$strip>]>>;
 }, z.core.$strip>;
-export type LlmConfigInput = z.infer<typeof llmConfigBaseSchema>;
-export declare const LLM_CONFIG_DEFAULTS: {
-    readonly ctx_size: 1024;
-    readonly gpu_layers: 99;
-    readonly device: "gpu";
-    readonly system_prompt: "You are a helpful assistant.";
-};
-export declare const llmConfigSchema: z.ZodPipe<z.ZodObject<{
-    ctx_size: z.ZodOptional<z.ZodNumber>;
-    temp: z.ZodOptional<z.ZodNumber>;
-    top_p: z.ZodOptional<z.ZodNumber>;
-    top_k: z.ZodOptional<z.ZodNumber>;
-    seed: z.ZodOptional<z.ZodNumber>;
-    gpu_layers: z.ZodOptional<z.ZodNumber>;
-    lora: z.ZodOptional<z.ZodString>;
-    device: z.ZodOptional<z.ZodString>;
-    predict: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<-1>, z.ZodLiteral<-2>, z.ZodNumber]>>;
-    system_prompt: z.ZodOptional<z.ZodString>;
-    no_mmap: z.ZodOptional<z.ZodBoolean>;
-    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
-    presence_penalty: z.ZodOptional<z.ZodNumber>;
-    frequency_penalty: z.ZodOptional<z.ZodNumber>;
-    repeat_penalty: z.ZodOptional<z.ZodNumber>;
-    stop_sequences: z.ZodOptional<z.ZodArray<z.ZodString>>;
-    n_discarded: z.ZodOptional<z.ZodNumber>;
-    tools: z.ZodOptional<z.ZodBoolean>;
-    projectionModelSrc: z.ZodOptional<z.ZodUnion<readonly [z.ZodString, z.ZodObject<{
-        src: z.ZodString;
-        name: z.ZodOptional<z.ZodString>;
-        modelId: z.ZodOptional<z.ZodString>;
-        registryPath: z.ZodOptional<z.ZodString>;
-        registrySource: z.ZodOptional<z.ZodString>;
-        blobCoreKey: z.ZodOptional<z.ZodString>;
-        blobIndex: z.ZodOptional<z.ZodNumber>;
-        engine: z.ZodOptional<z.ZodString>;
-        expectedSize: z.ZodOptional<z.ZodNumber>;
-        sha256Checksum: z.ZodOptional<z.ZodString>;
-        addon: z.ZodOptional<z.ZodUnion<readonly [z.ZodEnum<{
-            "llamacpp-completion": "llamacpp-completion";
-            "whispercpp-transcription": "whispercpp-transcription";
-            "llamacpp-embedding": "llamacpp-embedding";
-            "nmtcpp-translation": "nmtcpp-translation";
-            "onnx-tts": "onnx-tts";
-            "parakeet-transcription": "parakeet-transcription";
-            "onnx-ocr": "onnx-ocr";
-            llm: "llm";
-            whisper: "whisper";
-            embeddings: "embeddings";
-            nmt: "nmt";
-            parakeet: "parakeet";
-            tts: "tts";
-            ocr: "ocr";
-        }>, z.ZodLiteral<"vad">]>>;
-    }, z.core.$strip>]>>;
-}, z.core.$strip>, z.ZodTransform<{
-    ctx_size: number | undefined;
-    temp?: number | undefined;
-    top_p?: number | undefined;
-    top_k?: number | undefined;
-    seed?: number | undefined;
-    gpu_layers: number | undefined;
-    lora?: string | undefined;
-    device: string | undefined;
-    predict?: number | undefined;
-    system_prompt: string | undefined;
-    no_mmap?: boolean | undefined;
-    verbosity?: 0 | 3 | 2 | 1 | undefined;
-    presence_penalty?: number | undefined;
-    frequency_penalty?: number | undefined;
-    repeat_penalty?: number | undefined;
-    stop_sequences?: string[] | undefined;
-    n_discarded?: number | undefined;
-    tools?: boolean | undefined;
-    projectionModelSrc?: string | {
-        src: string;
-        name?: string | undefined;
-        modelId?: string | undefined;
-        registryPath?: string | undefined;
-        registrySource?: string | undefined;
-        blobCoreKey?: string | undefined;
-        blobIndex?: number | undefined;
-        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;
-    } | undefined;
-}, {
-    ctx_size?: number | undefined;
-    temp?: number | undefined;
-    top_p?: number | undefined;
-    top_k?: number | undefined;
-    seed?: number | undefined;
-    gpu_layers?: number | undefined;
-    lora?: string | undefined;
-    device?: string | undefined;
-    predict?: number | undefined;
-    system_prompt?: string | undefined;
-    no_mmap?: boolean | undefined;
-    verbosity?: 0 | 3 | 2 | 1 | undefined;
-    presence_penalty?: number | undefined;
-    frequency_penalty?: number | undefined;
-    repeat_penalty?: number | undefined;
-    stop_sequences?: string[] | undefined;
-    n_discarded?: number | undefined;
-    tools?: boolean | undefined;
-    projectionModelSrc?: string | {
-        src: string;
-        name?: string | undefined;
-        modelId?: string | undefined;
-        registryPath?: string | undefined;
-        registrySource?: string | undefined;
-        blobCoreKey?: string | undefined;
-        blobIndex?: number | undefined;
-        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;
-    } | undefined;
-}>>;
-export type LlmConfig = z.infer<typeof llmConfigSchema>;
-export declare const embedConfigBaseSchema: z.ZodObject<{
-    gpuLayers: z.ZodOptional<z.ZodNumber>;
-    device: z.ZodOptional<z.ZodEnum<{
-        gpu: "gpu";
-        cpu: "cpu";
-    }>>;
-    batchSize: z.ZodOptional<z.ZodNumber>;
-    pooling: z.ZodOptional<z.ZodEnum<{
-        none: "none";
-        mean: "mean";
-        cls: "cls";
-        last: "last";
-        rank: "rank";
-    }>>;
-    attention: z.ZodOptional<z.ZodEnum<{
-        causal: "causal";
-        "non-causal": "non-causal";
-    }>>;
-    embdNormalize: z.ZodOptional<z.ZodNumber>;
-    flashAttention: z.ZodOptional<z.ZodEnum<{
-        on: "on";
-        off: "off";
-        auto: "auto";
-    }>>;
-    mainGpu: z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
-        integrated: "integrated";
-        dedicated: "dedicated";
-    }>]>>;
-    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
+export declare const embedConfigSchema: z.ZodObject<{
+    config: z.ZodDefault<z.ZodString>;
 }, z.core.$strip>;
-export type EmbedConfigInput = z.infer<typeof embedConfigBaseSchema>;
-export declare const EMBED_CONFIG_DEFAULTS: {
-    readonly gpuLayers: 99;
-    readonly device: "gpu";
-    readonly batchSize: 1024;
-};
-export declare const embedConfigSchema: z.ZodPipe<z.ZodObject<{
-    gpuLayers: z.ZodOptional<z.ZodNumber>;
-    device: z.ZodOptional<z.ZodEnum<{
-        gpu: "gpu";
-        cpu: "cpu";
-    }>>;
-    batchSize: z.ZodOptional<z.ZodNumber>;
-    pooling: z.ZodOptional<z.ZodEnum<{
-        none: "none";
-        mean: "mean";
-        cls: "cls";
-        last: "last";
-        rank: "rank";
-    }>>;
-    attention: z.ZodOptional<z.ZodEnum<{
-        causal: "causal";
-        "non-causal": "non-causal";
-    }>>;
-    embdNormalize: z.ZodOptional<z.ZodNumber>;
-    flashAttention: z.ZodOptional<z.ZodEnum<{
-        on: "on";
-        off: "off";
-        auto: "auto";
-    }>>;
-    mainGpu: z.ZodOptional<z.ZodUnion<readonly [z.ZodNumber, z.ZodEnum<{
-        integrated: "integrated";
-        dedicated: "dedicated";
-    }>]>>;
-    verbosity: z.ZodOptional<z.ZodUnion<readonly [z.ZodLiteral<0>, z.ZodLiteral<1>, z.ZodLiteral<2>, z.ZodLiteral<3>]>>;
-}, z.core.$strip>, z.ZodTransform<{
-    gpuLayers: number | undefined;
-    device: "gpu" | "cpu" | undefined;
-    batchSize: number | undefined;
-    pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
-    attention?: "causal" | "non-causal" | undefined;
-    embdNormalize?: number | undefined;
-    flashAttention?: "on" | "off" | "auto" | undefined;
-    mainGpu?: number | "integrated" | "dedicated" | undefined;
-    verbosity?: 0 | 3 | 2 | 1 | undefined;
-}, {
-    gpuLayers?: number | undefined;
-    device?: "gpu" | "cpu" | undefined;
-    batchSize?: number | undefined;
-    pooling?: "none" | "mean" | "cls" | "last" | "rank" | undefined;
-    attention?: "causal" | "non-causal" | undefined;
-    embdNormalize?: number | undefined;
-    flashAttention?: "on" | "off" | "auto" | undefined;
-    mainGpu?: number | "integrated" | "dedicated" | undefined;
-    verbosity?: 0 | 3 | 2 | 1 | undefined;
-}>>;
+export type LlmConfig = z.infer<typeof llmConfigSchema>;
 export type EmbedConfig = z.infer<typeof embedConfigSchema>;
 //# sourceMappingURL=llamacpp-config.d.ts.map

package/dist/schemas/llamacpp-config.js

@@ -1,27 +1,23 @@
 import { z } from "zod";
-import { modelSrcInputSchema } from "./model-src-utils.js";
 export const VERBOSITY = {
-    ERROR: 0,
-    WARN: 1,
-    INFO: 2,
-    DEBUG: 3,
+    SILENT: -1,
+    INFO: 0,
+    DEBUG: 1,
 };
 const verbositySchema = z.union([
-    z.literal(VERBOSITY.ERROR),
-    z.literal(VERBOSITY.WARN),
+    z.literal(VERBOSITY.SILENT),
     z.literal(VERBOSITY.INFO),
     z.literal(VERBOSITY.DEBUG),
 ]);
-// Base schema - validates types, all fields optional (for client-side validation)
-export const llmConfigBaseSchema = z.object({
-    ctx_size: z.number().optional(),
+export const llmConfigSchema = z.object({
+    ctx_size: z.number().default(1024),
     temp: z.number().min(0).max(2).optional(),
     top_p: z.number().min(0).max(1).optional(),
     top_k: z.number().int().min(0).max(128).optional(),
     seed: z.number().optional(),
-    gpu_layers: z.number().optional(),
+    gpu_layers: z.number().default(99),
     lora: z.string().optional(),
-    device: z.string().optional(),
+    device: z.string().default("gpu"),
     predict: z
         .union([
         z.literal(-1), // special: until stop token
@@ -29,7 +25,7 @@ export const llmConfigBaseSchema = z.object({
         z.number().int().min(1), // positive integer: fixed token count
     ])
         .optional(),
-    system_prompt: z.string().optional(),
+    system_prompt: z.string().default("You are a helpful assistant."),
     no_mmap: z.boolean().optional(),
     verbosity: verbositySchema.optional(),
     presence_penalty: z.number().optional(),
@@ -38,42 +34,10 @@ export const llmConfigBaseSchema = z.object({
     stop_sequences: z.array(z.string()).optional(),
     n_discarded: z.number().optional(),
     tools: z.boolean().optional(),
-    projectionModelSrc: modelSrcInputSchema.optional(),
 });
-// Default values - typed as partial of the config
-export const LLM_CONFIG_DEFAULTS = {
-    ctx_size: 1024,
-    gpu_layers: 99,
-    device: "gpu",
-    system_prompt: "You are a helpful assistant.",
-};
-// Full schema - applies defaults via transform (no duplication)
-export const llmConfigSchema = llmConfigBaseSchema.transform((data) => ({
-    ...LLM_CONFIG_DEFAULTS,
-    ...data,
-}));
-// Base schema - validates types, all fields optional (for client-side validation)
-export const embedConfigBaseSchema = z.object({
-    gpuLayers: z.number().int().optional(),
-    device: z.enum(["gpu", "cpu"]).optional(),
-    batchSize: z.number().int().min(1).optional(),
-    pooling: z.enum(["none", "mean", "cls", "last", "rank"]).optional(),
-    attention: z.enum(["causal", "non-causal"]).optional(),
-    embdNormalize: z.number().int().optional(),
-    flashAttention: z.enum(["on", "off", "auto"]).optional(),
-    mainGpu: z
-        .union([z.number().int().min(0), z.enum(["integrated", "dedicated"])])
-        .optional(),
-    verbosity: verbositySchema.optional(),
+export const embedConfigSchema = z.object({
+    // https://github.com/tetherto/qvac-lib-infer-llamacpp-embed/tree/main?tab=readme-ov-file#4-create-config
+    config: z
+        .string()
+        .default("-ngl\t99\n-dev\tgpu\n--embd-separator\t⟪§§§EMBED_SEP§§§⟫\n-c\t512"),
 });
-// Default values - typed as partial of the config
-export const EMBED_CONFIG_DEFAULTS = {
-    gpuLayers: 99,
-    device: "gpu",
-    batchSize: 1024,
-};
-// Full schema - validates then applies defaults via transform
-export const embedConfigSchema = embedConfigBaseSchema.transform((data) => ({
-    ...EMBED_CONFIG_DEFAULTS,
-    ...data,
-}));