QVAC-17434 chore[bc]: migrate SDK plugins to new addon constructor shape

packages/sdk/examples/diffusion-flux2-klein.ts

@@ -33,6 +33,7 @@ const { progressStream, outputs, stats } = diffusion({
   height: 512,
   steps: 20,
   guidance: 3.5,
+  cfg_scale: 1,
   seed: -1,
 });
 

packages/sdk/examples/diffusion-txt2img.ts

@@ -27,7 +27,8 @@ const { progressStream, outputs, stats } = diffusion({
   width: 512,
   height: 512,
   steps: 20,
-  cfg_scale: 7.0,
+  guidance: 3.5,
+  cfg_scale: 1,
   seed: -1,
 });
 

packages/sdk/package.json

@@ -173,12 +173,11 @@
   },
   "dependencies": {
     "@qvac/decoder-audio": "^0.3.7",
-    "@qvac/diffusion-cpp": "^0.2.0",
-    "@qvac/dl-filesystem": "^0.2.1",
-    "@qvac/embed-llamacpp": "^0.13.4",
+    "@qvac/diffusion-cpp": "^0.3.0",
+    "@qvac/embed-llamacpp": "^0.14.0",
     "@qvac/error": "^0.1.1",
     "@qvac/langdetect-text": "^0.1.2",
-    "@qvac/llm-llamacpp": "^0.15.0",
+    "@qvac/llm-llamacpp": "^0.16.0",
     "@qvac/logging": "^0.1.0",
     "@qvac/ocr-onnx": "^0.4.2",
     "@qvac/rag": "^0.4.4",

packages/sdk/schemas/plugin.ts

@@ -83,7 +83,6 @@ export interface PluginModel {
 
 export interface PluginModelResult {
   model: PluginModel;
-  loader: unknown;
 }
 
 export interface PluginLogging {

packages/sdk/server/bare/ops/load-model.ts

@@ -27,7 +27,6 @@ import {
   ModelFileLocateFailedError,
 } from "@/utils/errors-server";
 import { getPlugin } from "@/server/plugins";
-import type FilesystemDL from "@qvac/dl-filesystem";
 import { promises as fsPromises } from "bare-fs";
 import path from "bare-path";
 import { getServerLogger } from "@/logging";
@@ -107,7 +106,7 @@ export async function loadModel(
       modelConfig: modelConfig as Record<string, unknown>,
       modelName,
       artifacts,
-    }) as { model: AnyModel; loader: FilesystemDL };
+    }) as { model: AnyModel };
 
     await result.model.load(false);
 
@@ -123,7 +122,6 @@ export async function loadModel(
       config: modelConfig,
       modelType: modelType as CanonicalModelType,
       name: modelName,
-      loader: result.loader,
     });
 
     return modelInitializationTimeMs !== undefined

packages/sdk/server/bare/ops/unload-model.ts

@@ -21,10 +21,6 @@ export async function unloadModel(params: UnloadModelParams) {
 
   clearFinetuneRuntimeState(modelId);
 
-  if (entry.local?.loader) {
-    await entry.local.loader.close();
-  }
-
   if (entry.local?.model && entry.local.model.unload) {
     await entry.local.model.unload();
   }

packages/sdk/server/bare/plugins/llamacpp-completion/plugin.ts

@@ -1,4 +1,4 @@
-import LlmLlamacpp, { type Loader as LlmLoader } from "@qvac/llm-llamacpp";
+import LlmLlamacpp from "@qvac/llm-llamacpp";
 import llmAddonLogging from "@qvac/llm-llamacpp/addonLogging";
 import {
   definePlugin,
@@ -21,9 +21,7 @@ import {
   type LlmConfigInput,
 } from "@/schemas";
 import { createStreamLogger, registerAddonLogger } from "@/logging";
-import { parseModelPath } from "@/server/utils";
-import FilesystemDL from "@qvac/dl-filesystem";
-import { asLoader } from "@/server/bare/utils/loader-adapter";
+import { expandGGUFIntoShards } from "@/server/utils";
 import { completion } from "@/server/bare/plugins/llamacpp-completion/ops/completion-stream";
 import { finetune } from "@/server/bare/plugins/llamacpp-completion/ops/finetune";
 import { translate } from "@/server/bare/ops/translate";
@@ -69,28 +67,22 @@ function createLlmModel(
   llmConfig: LlmConfig,
   projectionModelPath?: string,
 ) {
-  const { dirPath, basePath } = parseModelPath(modelPath);
-  const loader = new FilesystemDL({ dirPath });
   const logger = createStreamLogger(modelId, ModelType.llamacppCompletion);
   registerAddonLogger(modelId, ModelType.llamacppCompletion, logger);
   const llmConfigStrings = transformLlmConfig(llmConfig);
-
-  const args = {
-    loader: asLoader<LlmLoader>(loader),
-    opts: { stats: true },
+  const modelFiles = expandGGUFIntoShards(modelPath);
+
+  const model = new LlmLlamacpp({
+    files: {
+      model: modelFiles,
+      ...(projectionModelPath && { projectionModel: projectionModelPath }),
+    },
+    config: llmConfigStrings,
     logger,
-    diskPath: dirPath,
-    modelName: basePath,
-    projectionModel: projectionModelPath
-      ? parseModelPath(projectionModelPath).basePath
-      : "",
-    modelPath,
-    modelConfig: llmConfigStrings,
-  };
-
-  const model = new LlmLlamacpp(args, llmConfigStrings);
-
-  return { model, loader };
+    opts: { stats: true },
+  });
+
+  return { model };
 }
 
 export const llmPlugin = definePlugin({
@@ -116,14 +108,14 @@ export const llmPlugin = definePlugin({
   createModel(params: CreateModelParams): PluginModelResult {
     const llmConfig = (params.modelConfig ?? {}) as LlmConfig;
 
-    const { model, loader } = createLlmModel(
+    const { model } = createLlmModel(
       params.modelId,
       params.modelPath,
       llmConfig,
       params.artifacts?.["projectionModelPath"],
     );
 
-    return { model, loader };
+    return { model };
   },
 
   handlers: {

packages/sdk/server/bare/plugins/llamacpp-embedding/plugin.ts

@@ -1,7 +1,4 @@
-import EmbedLlamacpp, {
-  type GGMLConfig,
-  type Loader as EmbedLoader,
-} from "@qvac/embed-llamacpp";
+import EmbedLlamacpp, { type GGMLConfig } from "@qvac/embed-llamacpp";
 import embedAddonLogging from "@qvac/embed-llamacpp/addonLogging";
 import {
   definePlugin,
@@ -16,9 +13,7 @@ import {
   type EmbedConfig,
 } from "@/schemas";
 import { createStreamLogger, registerAddonLogger } from "@/logging";
-import { parseModelPath } from "@/server/utils";
-import FilesystemDL from "@qvac/dl-filesystem";
-import { asLoader } from "@/server/bare/utils/loader-adapter";
+import { expandGGUFIntoShards } from "@/server/utils";
 import { embed } from "@/server/bare/ops/embed";
 import { forwardModelExecution } from "@/profiling/model-execution";
 
@@ -68,25 +63,20 @@ function createEmbeddingsModel(
   modelPath: string,
   embedConfig: EmbedConfig,
 ) {
-  const { dirPath, basePath } = parseModelPath(modelPath);
-  const loader = new FilesystemDL({ dirPath });
   const logger = createStreamLogger(modelId, ModelType.llamacppEmbedding);
   registerAddonLogger(modelId, ModelType.llamacppEmbedding, logger);
 
   const config = transformEmbedConfig(embedConfig);
+  const modelFiles = expandGGUFIntoShards(modelPath);
 
-  const args = {
-    loader: asLoader<EmbedLoader>(loader),
-    opts: { stats: true },
+  const model = new EmbedLlamacpp({
+    files: { model: modelFiles },
+    config,
     logger,
-    diskPath: dirPath,
-    modelName: basePath,
-    modelPath,
-  };
-
-  const model = new EmbedLlamacpp(args, config);
+    opts: { stats: true },
+  });
 
-  return { model, loader };
+  return { model };
 }
 
 export const embeddingsPlugin = definePlugin({
@@ -98,13 +88,13 @@ export const embeddingsPlugin = definePlugin({
   createModel(params: CreateModelParams): PluginModelResult {
     const embedConfig = (params.modelConfig ?? {}) as EmbedConfig;
 
-    const { model, loader } = createEmbeddingsModel(
+    const { model } = createEmbeddingsModel(
       params.modelId,
       params.modelPath,
       embedConfig,
     );
 
-    return { model, loader };
+    return { model };
   },
 
   handlers: {

packages/sdk/server/bare/plugins/nmtcpp-translation/plugin.ts

@@ -20,7 +20,6 @@ import {
   type ResolveResult,
 } from "@/schemas";
 import { createStreamLogger, registerAddonLogger } from "@/logging";
-import { parseModelPath } from "@/server/utils";
 import path from "bare-path";
 import { ModelLoadFailedError } from "@/utils/errors-server";
 import { translate } from "@/server/bare/ops/translate";
@@ -73,7 +72,8 @@ function deriveBergamotRegistryVocabSources(modelSrc: string) {
  * Returns null if modelPath is not a recognisable Bergamot model.
  */
 function deriveColocatedBergamotVocabPaths(modelPath: string) {
-  const { dirPath, basePath } = parseModelPath(modelPath);
+  const dirPath = path.dirname(modelPath);
+  const basePath = path.basename(modelPath);
   const match = basePath.match(/^model\.([a-z]+)\.intgemm\.alphas\.bin$/);
   if (!match?.[1]) return null;
 
@@ -162,7 +162,7 @@ function createNmtModel(
     opts: { stats: true },
   });
 
-  return { model, loader: null };
+  return { model };
 }
 
 async function resolveBergamotVocab(
@@ -286,7 +286,7 @@ export const nmtPlugin = definePlugin({
     const pivotSrcVocabPath = artifacts["pivotSrcVocabPath"] ?? pivotDerived?.srcVocabPath;
     const pivotDstVocabPath = artifacts["pivotDstVocabPath"] ?? pivotDerived?.dstVocabPath;
 
-    const { model, loader } = createNmtModel(
+    const { model } = createNmtModel(
       params.modelId,
       params.modelPath,
       nmtConfig,
@@ -297,7 +297,7 @@ export const nmtPlugin = definePlugin({
       pivotDstVocabPath,
     );
 
-    return { model, loader };
+    return { model };
   },
 
   handlers: {

packages/sdk/server/bare/plugins/onnx-ocr/plugin.ts

@@ -127,7 +127,7 @@ export const ocrPlugin = definePlugin({
         ocrConfig,
     );
 
-    return { model, loader: null };
+    return { model };
   },
 
   handlers: {

packages/sdk/server/bare/plugins/onnx-tts/plugin.ts

@@ -199,7 +199,7 @@ function createChatterboxModel(
     opts: { stats: true },
     exclusiveRun: true,
   } as never);
-  return { model, loader: undefined };
+  return { model };
 }
 
 function createSupertonicModel(
@@ -250,7 +250,7 @@ function createSupertonicModel(
     opts: { stats: true },
     exclusiveRun: true,
   } as never);
-  return { model, loader: undefined };
+  return { model };
 }
 
 export const ttsPlugin = definePlugin({

packages/sdk/server/bare/plugins/parakeet-transcription/plugin.ts

@@ -19,12 +19,10 @@ import {
   type ResolveResult,
 } from "@/schemas";
 import { createStreamLogger, registerAddonLogger } from "@/logging";
-import { parseModelPath } from "@/server/utils";
 import {
   ModelLoadFailedError,
   ParakeetArtifactsRequiredError,
 } from "@/utils/errors-server";
-import FilesystemDL from "@qvac/dl-filesystem";
 import { transcribe } from "@/server/bare/ops/transcribe";
 import { attachModelExecutionMs } from "@/profiling/model-execution";
 
@@ -169,8 +167,6 @@ function createParakeetModel(
     );
   }
 
-  const { dirPath } = parseModelPath(primaryPath);
-  const loader = new FilesystemDL({ dirPath });
   const logger = createStreamLogger(params.modelId, ModelType.parakeetTranscription);
   registerAddonLogger(params.modelId, ModelType.parakeetTranscription, logger);
 
@@ -198,7 +194,7 @@ function createParakeetModel(
     logger,
   });
 
-  return { model, loader };
+  return { model };
 }
 
 export const parakeetPlugin = definePlugin({

packages/sdk/server/bare/plugins/sdcpp-generation/plugin.ts

@@ -1,4 +1,7 @@
-import ImgStableDiffusion, { type ImgStableDiffusionArgs, type SdConfig } from "@qvac/diffusion-cpp";
+import ImgStableDiffusion, {
+  type DiffusionFiles,
+  type SdConfig,
+} from "@qvac/diffusion-cpp";
 import addonLogging from "@qvac/diffusion-cpp/addonLogging";
 import {
   definePlugin,
@@ -15,7 +18,6 @@ import {
   type SdcppConfig,
 } from "@/schemas";
 import { createStreamLogger, registerAddonLogger } from "@/logging";
-import { parseModelPath } from "@/server/utils";
 import { diffusion } from "./ops/diffusion";
 
 type DiffusionArtifactKey =
@@ -72,25 +74,26 @@ export const diffusionPlugin = definePlugin({
   createModel(params: CreateModelParams): PluginModelResult {
     const { modelId, modelPath, modelConfig, artifacts } = params;
     const config = (modelConfig ?? {}) as SdcppConfig;
-    const { dirPath, basePath } = parseModelPath(modelPath);
     const logger = createStreamLogger(modelId, ModelType.sdcppGeneration);
     registerAddonLogger(modelId, ModelType.sdcppGeneration, logger);
 
-    const addonArgs: ImgStableDiffusionArgs = {
-      diskPath: dirPath,
-      modelName: basePath,
-      logger,
-      opts: { stats: true },
-      ...(artifacts?.["clipLModelPath"] && { clipLModel: artifacts["clipLModelPath"] }),
-      ...(artifacts?.["clipGModelPath"] && { clipGModel: artifacts["clipGModelPath"] }),
-      ...(artifacts?.["t5XxlModelPath"] && { t5XxlModel: artifacts["t5XxlModelPath"] }),
-      ...(artifacts?.["llmModelPath"] && { llmModel: artifacts["llmModelPath"] }),
-      ...(artifacts?.["vaeModelPath"] && { vaeModel: artifacts["vaeModelPath"] }),
+    const files: DiffusionFiles = {
+      model: modelPath,
+      ...(artifacts?.["clipLModelPath"] && { clipL: artifacts["clipLModelPath"] }),
+      ...(artifacts?.["clipGModelPath"] && { clipG: artifacts["clipGModelPath"] }),
+      ...(artifacts?.["t5XxlModelPath"] && { t5Xxl: artifacts["t5XxlModelPath"] }),
+      ...(artifacts?.["llmModelPath"] && { llm: artifacts["llmModelPath"] }),
+      ...(artifacts?.["vaeModelPath"] && { vae: artifacts["vaeModelPath"] }),
     };
 
-    const model = new ImgStableDiffusion(addonArgs, config as SdConfig);
+    const model = new ImgStableDiffusion({
+      files,
+      config: config as SdConfig,
+      logger,
+      opts: { stats: true },
+    });
 
-    return { model, loader: undefined };
+    return { model };
   },
 
   handlers: {