@qvac/sdk 0.16.0 → 0.17.0

package/dist/server/bare/plugins/llamacpp-completion/plugin.d.ts

@@ -1,4 +1,4 @@
-import { type CreateModelParams, type PluginModelResult, type ResolveContext, type LlmConfigInput } from '../../../../schemas/index';
+import { type CreateModelParams, type PluginModelResult, type ResolveContext, type LlmConfigInput } from '../../../../schemas/index.js';
 export declare const llmPlugin: {
     modelType: "llamacpp-completion";
     displayName: string;
@@ -60,6 +60,7 @@ export declare const llmPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -73,6 +74,7 @@ export declare const llmPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -161,7 +163,7 @@ export declare const llmPlugin: {
     }>;
     createModel(params: CreateModelParams): PluginModelResult;
     handlers: {
-        batchCompletionStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        batchCompletionStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             stream: import("zod").ZodOptional<import("zod").ZodBoolean>;
             captureThinking: import("zod").ZodOptional<import("zod").ZodBoolean>;
@@ -173,6 +175,7 @@ export declare const llmPlugin: {
                 harmony: "harmony";
                 qwen35: "qwen35";
                 gemma4: "gemma4";
+                dsml: "dsml";
             }>>;
             requestId: import("zod").ZodOptional<import("zod").ZodString>;
             prompts: import("zod").ZodArray<import("zod").ZodObject<{
@@ -280,6 +283,7 @@ export declare const llmPlugin: {
                         cacheTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                         promptTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                         generatedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
+                        emittedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                         avgConcurrentSeq: import("zod").ZodOptional<import("zod").ZodNumber>;
                         backendDevice: import("zod").ZodOptional<import("zod").ZodEnum<{
                             gpu: "gpu";
@@ -316,6 +320,7 @@ export declare const llmPlugin: {
                 cacheTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                 promptTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                 generatedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
+                emittedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                 avgConcurrentSeq: import("zod").ZodOptional<import("zod").ZodNumber>;
                 backendDevice: import("zod").ZodOptional<import("zod").ZodEnum<{
                     gpu: "gpu";
@@ -323,7 +328,7 @@ export declare const llmPlugin: {
                 }>>;
             }, import("zod/v4/core").$strip>>;
         }, import("zod/v4/core").$strict>>;
-        completionStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        completionStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             history: import("zod").ZodArray<import("zod").ZodObject<{
                 role: import("zod").ZodString;
                 content: import("zod").ZodString;
@@ -376,6 +381,7 @@ export declare const llmPlugin: {
                 harmony: "harmony";
                 qwen35: "qwen35";
                 gemma4: "gemma4";
+                dsml: "dsml";
             }>>;
             responseFormat: import("zod").ZodOptional<import("zod").ZodDiscriminatedUnion<[import("zod").ZodObject<{
                 type: import("zod").ZodLiteral<"text">;
@@ -437,6 +443,7 @@ export declare const llmPlugin: {
                     cacheTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                     promptTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                     generatedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
+                    emittedTokens: import("zod").ZodOptional<import("zod").ZodNumber>;
                     avgConcurrentSeq: import("zod").ZodOptional<import("zod").ZodNumber>;
                     backendDevice: import("zod").ZodOptional<import("zod").ZodEnum<{
                         gpu: "gpu";
@@ -467,7 +474,7 @@ export declare const llmPlugin: {
                 }, import("zod/v4/core").$strip>>;
             }, import("zod/v4/core").$strict>]>>;
         }, import("zod/v4/core").$strict>>;
-        finetune: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodUnion<readonly [import("zod").ZodObject<{
+        finetune: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodUnion<readonly [import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             options: import("zod").ZodObject<{
                 trainDatasetDir: import("zod").ZodString;
@@ -618,7 +625,7 @@ export declare const llmPlugin: {
                 epochs_completed: import("zod").ZodNumber;
             }, import("zod/v4/core").$strict>>;
         }, import("zod/v4/core").$strict>>;
-        translate: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodDiscriminatedUnion<[import("zod").ZodObject<{
+        translate: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodDiscriminatedUnion<[import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             text: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodArray<import("zod").ZodString>]>;
             stream: import("zod").ZodBoolean;

package/dist/server/bare/plugins/llamacpp-completion/plugin.js

@@ -51,27 +51,27 @@ var __disposeResources = (this && this.__disposeResources) || (function (Suppres
     return e.name = "SuppressedError", e.error = error, e.suppressed = suppressed, e;
 });
 import LlmLlamacpp from '@qvac/llm-llamacpp';
-import llmAddonLogging from '@qvac/llm-llamacpp/addonLogging.js';
+import llmAddonLogging from '@qvac/llm-llamacpp/addonLogging';
 import { definePlugin, defineHandler, finetuneRequestSchema, batchCompletionStreamRequestSchema, batchCompletionStreamResponseSchema, completionStreamRequestSchema, completionStreamResponseSchema, finetuneResponseSchema, translateRequestSchema, translateResponseSchema, ModelType, llmConfigBaseSchema, ADDON_LLM } from '../../../../schemas/index.js';
 import { createStreamLogger, registerAddonLogger } from '../../../../logging/index.js';
-import { expandGGUFIntoShards } from '../../../utils/index.js';
-import { completion } from './ops/completion-stream.js';
-import { batchCompletion } from './ops/batch-completion-stream.js';
-import { finetune } from './ops/finetune.js';
-import { translate } from '../../ops/translate.js';
-import { transformLlmConfig } from './transform.js';
+import { expandGGUFIntoShards } from '../../../../server/utils/index.js';
+import { completion } from '../../../../server/bare/plugins/llamacpp-completion/ops/completion-stream.js';
+import { batchCompletion } from '../../../../server/bare/plugins/llamacpp-completion/ops/batch-completion-stream.js';
+import { finetune } from '../../../../server/bare/plugins/llamacpp-completion/ops/finetune.js';
+import { translate } from '../../../../server/bare/ops/translate.js';
+import { transformLlmConfig } from '../../../../server/bare/plugins/llamacpp-completion/transform.js';
 import { attachModelExecutionMs } from '../../../../profiling/model-execution.js';
-import { getModelConfig } from '../../registry/model-registry.js';
-import { createCompletionNormalizer } from '../../../utils/completion-normalizer.js';
-import { detectToolDialect } from '../../../utils/tool-integration.js';
-import { getRequestRegistry, withRequestContext } from '../../runtime/index.js';
-import { generateServerRequestId } from '../../runtime/request-id.js';
+import { getModelConfig } from '../../../../server/bare/registry/model-registry.js';
+import { createCompletionNormalizer } from '../../../../server/utils/completion-normalizer.js';
+import { detectToolDialect } from '../../../../server/utils/tool-integration.js';
+import { getRequestRegistry, withRequestContext } from '../../../../server/bare/runtime/index.js';
+import { generateServerRequestId } from '../../../../server/bare/runtime/request-id.js';
 import { getServerLogger } from '../../../../logging/index.js';
 import { ContextOverflowError } from '../../../../utils/errors-server.js';
-import { isAddonContextOverflowError, parseContextOverflowMessage } from './ops/context-overflow.js';
-import { isAddonCancelledError } from './ops/batch-cancelled.js';
-import { isMobile } from '../../registry/runtime-context-registry.js';
-import { stripMultiGpuKeys } from '../../../utils/multi-gpu-mobile.js';
+import { isAddonContextOverflowError, parseContextOverflowMessage } from '../../../../server/bare/plugins/llamacpp-completion/ops/context-overflow.js';
+import { isAddonCancelledError } from '../../../../server/bare/plugins/llamacpp-completion/ops/batch-cancelled.js';
+import { isMobile } from '../../../../server/bare/registry/runtime-context-registry.js';
+import { stripMultiGpuKeys } from '../../../../server/utils/multi-gpu-mobile.js';
 function createLlmModel(modelId, modelPath, llmConfig, projectionModelPath) {
     const logger = createStreamLogger(modelId, ModelType.llamacppCompletion);
     registerAddonLogger(modelId, ModelType.llamacppCompletion, logger);

package/dist/server/bare/plugins/llamacpp-completion/transform.d.ts

@@ -1,4 +1,4 @@
-import { type LlmConfig } from '../../../../schemas/index';
+import { type LlmConfig } from '../../../../schemas/index.js';
 /**
  * Converts an LlmConfig into the flat string-keyed map the C++ addon expects.
  *

package/dist/server/bare/plugins/llamacpp-embedding/plugin.d.ts

@@ -1,4 +1,4 @@
-import { type CreateModelParams, type PluginModelResult } from '../../../../schemas/index';
+import { type CreateModelParams, type PluginModelResult } from '../../../../schemas/index.js';
 export declare const embeddingsPlugin: {
     modelType: "llamacpp-embedding";
     displayName: string;
@@ -47,7 +47,7 @@ export declare const embeddingsPlugin: {
     }, import("zod/v4/core").$strip>;
     createModel(params: CreateModelParams): PluginModelResult;
     handlers: {
-        embed: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        embed: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             text: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodArray<import("zod").ZodString>]>;
             type: import("zod").ZodLiteral<"embed">;

package/dist/server/bare/plugins/llamacpp-embedding/plugin.js

@@ -1,12 +1,12 @@
 import EmbedLlamacpp, {} from '@qvac/embed-llamacpp';
-import embedAddonLogging from '@qvac/embed-llamacpp/addonLogging.js';
+import embedAddonLogging from '@qvac/embed-llamacpp/addonLogging';
 import { definePlugin, defineHandler, embedRequestSchema, embedResponseSchema, ModelType, embedConfigBaseSchema, ADDON_EMBEDDING } from '../../../../schemas/index.js';
 import { createStreamLogger, registerAddonLogger, getServerLogger } from '../../../../logging/index.js';
-import { expandGGUFIntoShards } from '../../../utils/index.js';
-import { embed } from '../../ops/embed.js';
+import { expandGGUFIntoShards } from '../../../../server/utils/index.js';
+import { embed } from '../../../../server/bare/ops/embed.js';
 import { forwardModelExecution } from '../../../../profiling/model-execution.js';
-import { isMobile } from '../../registry/runtime-context-registry.js';
-import { stripMultiGpuKeys } from '../../../utils/multi-gpu-mobile.js';
+import { isMobile } from '../../../../server/bare/registry/runtime-context-registry.js';
+import { stripMultiGpuKeys } from '../../../../server/utils/multi-gpu-mobile.js';
 function transformEmbedConfig(embedConfig) {
     const config = {
         device: embedConfig.device,

package/dist/server/bare/plugins/nmtcpp-translation/plugin.d.ts

@@ -1,4 +1,4 @@
-import { type CreateModelParams, type PluginModelResult, type ResolveContext, type ResolveResult } from '../../../../schemas/index';
+import { type CreateModelParams, type PluginModelResult, type ResolveContext, type ResolveResult } from '../../../../schemas/index.js';
 export declare const nmtPlugin: {
     modelType: "nmtcpp-translation";
     displayName: string;
@@ -142,6 +142,7 @@ export declare const nmtPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -155,6 +156,7 @@ export declare const nmtPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -179,6 +181,7 @@ export declare const nmtPlugin: {
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                 embeddings: "embeddings";
                 diffusion: "diffusion";
+                audiogen: "audiogen";
                 tts: "tts";
                 ocr: "ocr";
                 vla: "vla";
@@ -192,6 +195,7 @@ export declare const nmtPlugin: {
                 "parakeet-transcription": "parakeet-transcription";
                 "ggml-ocr": "ggml-ocr";
                 "sdcpp-generation": "sdcpp-generation";
+                "audiogen-ggml": "audiogen-ggml";
                 "ggml-vla": "ggml-vla";
                 "ggml-classification": "ggml-classification";
                 llm: "llm";
@@ -229,6 +233,7 @@ export declare const nmtPlugin: {
                 addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                     embeddings: "embeddings";
                     diffusion: "diffusion";
+                    audiogen: "audiogen";
                     tts: "tts";
                     ocr: "ocr";
                     vla: "vla";
@@ -242,6 +247,7 @@ export declare const nmtPlugin: {
                     "parakeet-transcription": "parakeet-transcription";
                     "ggml-ocr": "ggml-ocr";
                     "sdcpp-generation": "sdcpp-generation";
+                    "audiogen-ggml": "audiogen-ggml";
                     "ggml-vla": "ggml-vla";
                     "ggml-classification": "ggml-classification";
                     llm: "llm";
@@ -266,6 +272,7 @@ export declare const nmtPlugin: {
                 addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                     embeddings: "embeddings";
                     diffusion: "diffusion";
+                    audiogen: "audiogen";
                     tts: "tts";
                     ocr: "ocr";
                     vla: "vla";
@@ -279,6 +286,7 @@ export declare const nmtPlugin: {
                     "parakeet-transcription": "parakeet-transcription";
                     "ggml-ocr": "ggml-ocr";
                     "sdcpp-generation": "sdcpp-generation";
+                    "audiogen-ggml": "audiogen-ggml";
                     "ggml-vla": "ggml-vla";
                     "ggml-classification": "ggml-classification";
                     llm: "llm";
@@ -303,6 +311,7 @@ export declare const nmtPlugin: {
                 addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
                     embeddings: "embeddings";
                     diffusion: "diffusion";
+                    audiogen: "audiogen";
                     tts: "tts";
                     ocr: "ocr";
                     vla: "vla";
@@ -316,6 +325,7 @@ export declare const nmtPlugin: {
                     "parakeet-transcription": "parakeet-transcription";
                     "ggml-ocr": "ggml-ocr";
                     "sdcpp-generation": "sdcpp-generation";
+                    "audiogen-ggml": "audiogen-ggml";
                     "ggml-vla": "ggml-vla";
                     "ggml-classification": "ggml-classification";
                     llm: "llm";
@@ -401,7 +411,7 @@ export declare const nmtPlugin: {
     resolveConfig(cfg: Record<string, unknown>, ctx: ResolveContext): Promise<ResolveResult<Record<string, unknown>>>;
     createModel(params: CreateModelParams): PluginModelResult;
     handlers: {
-        translate: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodDiscriminatedUnion<[import("zod").ZodObject<{
+        translate: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodDiscriminatedUnion<[import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             text: import("zod").ZodUnion<readonly [import("zod").ZodString, import("zod").ZodArray<import("zod").ZodString>]>;
             stream: import("zod").ZodBoolean;

package/dist/server/bare/plugins/nmtcpp-translation/plugin.js

@@ -1,11 +1,11 @@
-import nmtAddonLogging from '@qvac/translation-nmtcpp/addonLogging.js';
+import nmtAddonLogging from '@qvac/translation-nmtcpp/addonLogging';
 import TranslationNmtcpp, {} from '@qvac/translation-nmtcpp';
 import { definePlugin, defineHandler, translateRequestSchema, translateResponseSchema, ModelType, nmtConfigBaseSchema, ADDON_NMT, BERGAMOT_CJK_LANG_PAIRS } from '../../../../schemas/index.js';
 import { createStreamLogger, registerAddonLogger } from '../../../../logging/index.js';
 import path from 'bare-path';
-import { translate } from '../../ops/translate.js';
+import { translate } from '../../../../server/bare/ops/translate.js';
 import { attachModelExecutionMs } from '../../../../profiling/model-execution.js';
-import { resolveBergamotVocab } from './resolve-vocab.js';
+import { resolveBergamotVocab } from '../../../../server/bare/plugins/nmtcpp-translation/resolve-vocab.js';
 /**
  * Derive absolute vocab paths from a resolved Bergamot model path.
  * Works for both companion-set layout (colocated files) and any layout where

package/dist/server/bare/plugins/nmtcpp-translation/resolve-vocab.d.ts

@@ -1,4 +1,4 @@
-import { type ModelSrcInput, type NmtConfig, type ResolveContext, type ResolveResult } from '../../../../schemas/index';
+import { type ModelSrcInput, type NmtConfig, type ResolveContext, type ResolveResult } from '../../../../schemas/index.js';
 export interface PivotModelConfig {
     modelSrc: string;
     srcVocabSrc?: ModelSrcInput;

package/dist/server/bare/plugins/parakeet-transcription/plugin.d.ts

@@ -1,15 +1,15 @@
-import { type ParakeetConfig, type CreateModelParams, type PluginModelResult, type ResolveResult } from '../../../../schemas/index';
+import { type ParakeetConfig, type CreateModelParams, type PluginModelResult, type ResolveResult } from '../../../../schemas/index.js';
 export declare const parakeetPlugin: {
     modelType: "parakeet-transcription";
     displayName: string;
-    addonPackage: "@qvac/transcription-parakeet";
+    addonPackage: "@qvac/asr-ggml";
     loadConfigSchema: import("zod").ZodObject<{
         [x: string]: import("zod").ZodOptional<import("zod").ZodUnknown>;
     }, import("zod/v4/core").$strict>;
     resolveConfig(cfg: ParakeetConfig): Promise<ResolveResult<ParakeetConfig>>;
     createModel(params: CreateModelParams): PluginModelResult;
     handlers: {
-        transcribe: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        transcribe: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             prompt: import("zod").ZodOptional<import("zod").ZodString>;
             metadata: import("zod").ZodOptional<import("zod").ZodBoolean>;
@@ -62,7 +62,7 @@ export declare const parakeetPlugin: {
             }, import("zod/v4/core").$strip>], "source">>>;
             type: import("zod").ZodLiteral<"transcribe">;
         }, import("zod/v4/core").$strip>>;
-        transcribeStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+        transcribeStream: import("../../../../schemas/index.js").PluginHandlerDefinition<import("zod").ZodObject<{
             modelId: import("zod").ZodString;
             prompt: import("zod").ZodOptional<import("zod").ZodString>;
             metadata: import("zod").ZodOptional<import("zod").ZodBoolean>;
@@ -127,8 +127,8 @@ export declare const parakeetPlugin: {
         }, import("zod/v4/core").$strip>>;
     };
     logging: {
-        module: import("@qvac/transcription-parakeet/addonLogging").AddonLogging;
-        namespace: "parakeet-transcription";
+        module: import("@qvac/asr-ggml/addonLogging").AddonLogging;
+        namespace: "@qvac/asr-ggml";
     };
 };
 //# sourceMappingURL=plugin.d.ts.map

package/dist/server/bare/plugins/parakeet-transcription/plugin.js

@@ -1,10 +1,11 @@
-import parakeetAddonLogging from '@qvac/transcription-parakeet/addonLogging.js';
-import TranscriptionParakeet, {} from '@qvac/transcription-parakeet';
-import { definePlugin, defineHandler, defineDuplexHandler, transcribeRequestSchema, transcribeResponseSchema, transcribeStreamRequestSchema, transcribeStreamResponseSchema, ModelType, parakeetLoadConfigSchema, LEGACY_PARAKEET_ONNX_MODEL_CONFIG_FIELDS, ADDON_PARAKEET } from '../../../../schemas/index.js';
-import { createStreamLogger, registerAddonLogger } from '../../../../logging/index.js';
+import asrAddonLogging from '@qvac/asr-ggml/addonLogging';
+import ASRGgml from '@qvac/asr-ggml';
+import { definePlugin, defineHandler, defineDuplexHandler, transcribeRequestSchema, transcribeResponseSchema, transcribeStreamRequestSchema, transcribeStreamResponseSchema, ModelType, parakeetLoadConfigSchema, LEGACY_PARAKEET_ONNX_MODEL_CONFIG_FIELDS, ADDON_ASR } from '../../../../schemas/index.js';
 import { ModelLoadFailedError, TranscriptionFailedError, LegacyParakeetModelDeprecatedError } from '../../../../utils/errors-server.js';
-import { transcribe, transcribeStream } from '../../ops/transcribe.js';
+import { transcribe, transcribeStream } from '../../../../server/bare/ops/transcribe.js';
 import { attachModelExecutionMs } from '../../../../profiling/model-execution.js';
+import { buildParakeetEngineConfig } from '../../../../server/bare/plugins/asr-ggml/config.js';
+import { createAsrModelLogger } from '../../../../server/bare/plugins/asr-ggml/logging.js';
 function resolveParakeetConfig(cfg) {
     const cfgRecord = cfg;
     const legacyFields = LEGACY_PARAKEET_ONNX_MODEL_CONFIG_FIELDS.filter((name) => cfgRecord[name] !== undefined);
@@ -19,19 +20,11 @@ function createParakeetModel(params) {
     if (!modelPath) {
         throw new ModelLoadFailedError('Parakeet requires a GGUF model source');
     }
-    const logger = createStreamLogger(params.modelId, ModelType.parakeetTranscription);
-    registerAddonLogger(params.modelId, ModelType.parakeetTranscription, logger);
-    const files = {
-        model: modelPath
-    };
-    const parakeetConfig = Object.fromEntries(Object.entries(config).filter(([, value]) => value !== undefined));
-    const addonConfig = {
+    const logger = createAsrModelLogger(params.modelId, ModelType.parakeetTranscription);
+    const model = new ASRGgml({
+        files: { model: modelPath },
+        config: buildParakeetEngineConfig(config),
         enableStats: true,
-        parakeetConfig
-    };
-    const model = new TranscriptionParakeet({
-        files,
-        config: addonConfig,
         logger
     });
     return { model };
@@ -39,7 +32,7 @@ function createParakeetModel(params) {
 export const parakeetPlugin = definePlugin({
     modelType: ModelType.parakeetTranscription,
     displayName: 'Parakeet (NVIDIA NeMo GGML)',
-    addonPackage: ADDON_PARAKEET,
+    addonPackage: ADDON_ASR,
     loadConfigSchema: parakeetLoadConfigSchema,
     resolveConfig(cfg) {
         return resolveParakeetConfig(cfg);
@@ -131,7 +124,7 @@ export const parakeetPlugin = definePlugin({
         })
     },
     logging: {
-        module: parakeetAddonLogging,
-        namespace: ModelType.parakeetTranscription
+        module: asrAddonLogging,
+        namespace: ADDON_ASR
     }
 });

package/dist/server/bare/plugins/sdcpp-generation/ops/diffusion.d.ts

@@ -1,3 +1,3 @@
-import type { DiffusionRequest, DiffusionStreamResponse } from '../../../../../schemas/sdcpp-config';
+import type { DiffusionRequest, DiffusionStreamResponse } from '../../../../../schemas/sdcpp-config.js';
 export declare function diffusion(request: DiffusionRequest): AsyncGenerator<DiffusionStreamResponse>;
 //# sourceMappingURL=diffusion.d.ts.map

package/dist/server/bare/plugins/sdcpp-generation/ops/diffusion.js

@@ -1,4 +1,4 @@
-import { getModel } from '../../../registry/model-registry.js';
+import { getModel } from '../../../../../server/bare/registry/model-registry.js';
 export async function* diffusion(request) {
     const model = getModel(request.modelId);
     const init_image = request.init_image ? Buffer.from(request.init_image, 'base64') : undefined;

package/dist/server/bare/plugins/sdcpp-generation/ops/upscale.d.ts

@@ -1,3 +1,3 @@
-import type { UpscaleRequest, UpscaleStreamResponse } from '../../../../../schemas/sdcpp-config';
+import type { UpscaleRequest, UpscaleStreamResponse } from '../../../../../schemas/sdcpp-config.js';
 export declare function upscale(request: UpscaleRequest): AsyncGenerator<UpscaleStreamResponse>;
 //# sourceMappingURL=upscale.d.ts.map

package/dist/server/bare/plugins/sdcpp-generation/ops/upscale.js

@@ -1,5 +1,5 @@
 import { EsrganUpscaler } from '@qvac/diffusion-cpp';
-import { getModel, getModelEntry } from '../../../registry/model-registry.js';
+import { getModel, getModelEntry } from '../../../../../server/bare/registry/model-registry.js';
 import { ModelOperationNotSupportedError } from '../../../../../utils/errors-server.js';
 import { ModelType } from '../../../../../schemas/index.js';
 // The diffusion plugin instantiates `EsrganUpscaler` when the model is loaded

package/dist/server/bare/plugins/sdcpp-generation/ops/video.d.ts

@@ -1,5 +1,6 @@
 import { VideoStableDiffusion } from '@qvac/diffusion-cpp';
-import { type VideoRequest, type VideoStreamResponse } from '../../../../../schemas/sdcpp-config';
+import { type VideoRequest, type VideoStreamResponse } from '../../../../../schemas/sdcpp-config.js';
 export declare function markLtxVideoModel(model: VideoStableDiffusion): void;
+export declare function markMoeCapableVideoModel(model: VideoStableDiffusion): void;
 export declare function video(request: VideoRequest): AsyncGenerator<VideoStreamResponse>;
 //# sourceMappingURL=video.d.ts.map

package/dist/server/bare/plugins/sdcpp-generation/ops/video.js

@@ -52,19 +52,23 @@ var __disposeResources = (this && this.__disposeResources) || (function (Suppres
 });
 import { VideoStableDiffusion } from '@qvac/diffusion-cpp';
 import { getServerLogger } from '../../../../../logging/index.js';
-import { getModel, getModelEntry } from '../../../registry/model-registry.js';
-import { getRequestRegistry, withRequestContext } from '../../../runtime/index.js';
-import { generateServerRequestId } from '../../../runtime/request-id.js';
+import { getModel, getModelEntry } from '../../../../../server/bare/registry/model-registry.js';
+import { getRequestRegistry, withRequestContext } from '../../../../../server/bare/runtime/index.js';
+import { generateServerRequestId } from '../../../../../server/bare/runtime/request-id.js';
 import { ModelOperationNotSupportedError, PluginRequestValidationFailedError } from '../../../../../utils/errors-server.js';
 import { formatZodError } from '../../../../../utils/zod-error.js';
 import { ModelType } from '../../../../../schemas/index.js';
-import { ltxVideoRequestSchema } from '../../../../../schemas/sdcpp-config.js';
+import { ltxVideoRequestSchema, singleExpertVideoRequestSchema } from '../../../../../schemas/sdcpp-config.js';
 const ltxVideoModels = new WeakSet();
+const moeCapableVideoModels = new WeakSet();
 export function markLtxVideoModel(model) {
     ltxVideoModels.add(model);
 }
-function parseLtxVideoRequest(request) {
-    const result = ltxVideoRequestSchema.safeParse(request);
+export function markMoeCapableVideoModel(model) {
+    moeCapableVideoModels.add(model);
+}
+function parseVideoRequest(schema, request) {
+    const result = schema.safeParse(request);
     if (!result.success) {
         throw new PluginRequestValidationFailedError('videoStream', formatZodError(result.error));
     }
@@ -105,7 +109,9 @@ export async function* video(request) {
         const requestLogger = withRequestContext(getServerLogger(), ctx);
         const model = asVideoModel(getModel(request.modelId), request.modelId);
         if (ltxVideoModels.has(model))
-            parseLtxVideoRequest(request);
+            parseVideoRequest(ltxVideoRequestSchema, request);
+        if (!moeCapableVideoModels.has(model))
+            parseVideoRequest(singleExpertVideoRequestSchema, request);
         const onAbort = () => {
             model.cancel().catch((err) => {
                 requestLogger.warn(`[cancel] model.cancel() rejected during abort for modelId=${request.modelId}: ${err instanceof Error ? err.message : String(err)}`);