@aparte/provider-transformers 0.16.3 → 0.16.5

package/dist/runners/types.d.ts

@@ -0,0 +1,69 @@
+/**
+ * The runner contract — where the provider's plumbing stops and the model begins.
+ *
+ * A runner owns loading and calling a model; the provider owns everything around it
+ * (the worker, the queue, download progress, cancel, the stream into the transcript, the
+ * catalogue). Two runners ship with the package (`text-generation`, `image-text-to-text`),
+ * and an app can point `TransformersModelConfig.runner` at a module of its own that
+ * exports `createRunner` — the worker imports it.
+ *
+ * Sized for the hardest known consumer, not the easiest: a runner EMITS the full stream
+ * vocabulary (text, thinking, tool_use, done, error — not tokens), has a command lane for
+ * what is not a generation (swapping an adapter, warming something up), reports progress
+ * with a free-form `detail`, and can dispose. A research-grade provider that today
+ * carries its own worker fits here without giving up anything of its own.
+ */
+import type { AparteChatMessage, AparteStreamEvent } from '@aparte/core';
+/** The Transformers.js module, as the worker resolved it — a runner never imports its own copy. */
+export type TransformersModule = typeof import('@huggingface/transformers');
+/** ONNX dtype, or a per-part map (`{ decoder_model_merged: 'q4' }`). */
+export type Dtype = string | Record<string, string>;
+export type Device = 'webgpu' | 'wasm' | 'auto';
+/** The runners this package ships. Selected with `TransformersModelConfig.task`. */
+export type BuiltInRunner = 'text-generation' | 'image-text-to-text';
+export interface RunnerProgress {
+    status: 'downloading' | 'loading' | 'cached' | 'ready';
+    file?: string;
+    /** 0–100. */
+    progress?: number;
+    /** Anything a runner wants to say that the four statuses do not — an adapter name, a stage. */
+    detail?: unknown;
+}
+export interface GenerationOptions {
+    maxTokens?: number;
+    temperature?: number;
+    seed?: number;
+}
+/** What a runner is created with. */
+export interface RunnerContext {
+    transformers: TransformersModule;
+    modelId: string;
+    dtype?: Dtype;
+    device?: Device;
+    /** Download / load progress, forwarded to the page. */
+    progress(p: RunnerProgress): void;
+    /** A developer-facing warning (dropped content, an unsupported option). Said once per message. */
+    warn(message: string): void;
+}
+export interface RunnerGenerateInput {
+    /** The conversation, WITH its content parts — the runner decides what to keep. */
+    messages: AparteChatMessage[];
+    options: GenerationOptions;
+    /** The stream, in aparté's vocabulary. `done` and `error` are the host's to close on. */
+    emit(event: AparteStreamEvent): void;
+    /** Fired when the page cancels; wire it to the model's stop. */
+    signal: AbortSignal;
+}
+export interface TransformersRunner {
+    /** Resolve after the last emit. Emitting `done` is optional — the host closes the stream either way. */
+    generate(input: RunnerGenerateInput): Promise<void>;
+    /** The extended surface: anything that is not a generation, opaque to the provider. */
+    command?(name: string, payload: unknown): Promise<unknown>;
+    dispose?(): void | Promise<void>;
+}
+export type CreateRunner = (ctx: RunnerContext) => Promise<TransformersRunner>;
+/** What a runner module exports. */
+export interface RunnerModule {
+    createRunner: CreateRunner;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/runners/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/runners/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AAEzE,mGAAmG;AACnG,MAAM,MAAM,kBAAkB,GAAG,cAAc,2BAA2B,CAAC,CAAC;AAE5E,wEAAwE;AACxE,MAAM,MAAM,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;AACpD,MAAM,MAAM,MAAM,GAAG,QAAQ,GAAG,MAAM,GAAG,MAAM,CAAC;AAEhD,oFAAoF;AACpF,MAAM,MAAM,aAAa,GAAG,iBAAiB,GAAG,oBAAoB,CAAC;AAErE,MAAM,WAAW,cAAc;IAC3B,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,QAAQ,GAAG,OAAO,CAAC;IACvD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,aAAa;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,+FAA+F;IAC/F,MAAM,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,iBAAiB;IAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,qCAAqC;AACrC,MAAM,WAAW,aAAa;IAC1B,YAAY,EAAE,kBAAkB,CAAC;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,uDAAuD;IACvD,QAAQ,CAAC,CAAC,EAAE,cAAc,GAAG,IAAI,CAAC;IAClC,kGAAkG;IAClG,IAAI,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;CAC/B;AAED,MAAM,WAAW,mBAAmB;IAChC,kFAAkF;IAClF,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAC9B,OAAO,EAAE,iBAAiB,CAAC;IAC3B,yFAAyF;IACzF,IAAI,CAAC,KAAK,EAAE,iBAAiB,GAAG,IAAI,CAAC;IACrC,gEAAgE;IAChE,MAAM,EAAE,WAAW,CAAC;CACvB;AAED,MAAM,WAAW,kBAAkB;IAC/B,wGAAwG;IACxG,QAAQ,CAAC,KAAK,EAAE,mBAAmB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACpD,uFAAuF;IACvF,OAAO,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC3D,OAAO,CAAC,IAAI,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACpC;AAED,MAAM,MAAM,YAAY,GAAG,CAAC,GAAG,EAAE,aAAa,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAC;AAE/E,oCAAoC;AACpC,MAAM,WAAW,YAAY;IACzB,YAAY,EAAE,YAAY,CAAC;CAC9B"}

package/dist/worker-host.d.ts

@@ -0,0 +1,102 @@
+/**
+ * The worker's logic, as a function of its three seams.
+ *
+ * `worker.ts` is the two-line shell that binds this to `self`; everything it decides is
+ * here, with `post`, `loadTransformers` and `importRunner` injected — so the protocol is
+ * tested with fakes where no Worker and no model can load (`worker-host.test.ts`).
+ *
+ * The protocol (main ⇄ worker):
+ *
+ *   → init            { transformersUrl? }                      once, first
+ *   → prepare         { id, modelId, task?, runner?, dtype?, device? }
+ *   → generate        { id, modelId, messages, options, task?, runner?, dtype?, device? }
+ *   → cancel          { id }
+ *   → command         { id, modelId, name, payload, task?, runner?, dtype?, device? }
+ *   ← progress        { id, status, file?, progress?, detail? }
+ *   ← pipeline-ready  { modelId }                                a runner is up
+ *   ← prepare-error   { id, message }
+ *   ← gen-event       { id, event }                              text | thinking | tool_use
+ *   ← gen-done        { id, usage? }                             the host closes; `done` is never forwarded raw
+ *   ← gen-error       { id, message }                            emitted or thrown
+ *   ← warning         { message }                                once per distinct text
+ *   ← command-result  { id, result } | { id, error }
+ *
+ * One runner is resident at a time, keyed by model AND runner: a switch disposes the
+ * previous one first, which is the "one pipeline per tab" rule this package has always
+ * kept — a local model is gigabytes, and two of them resident is the failure to avoid.
+ */
+import type { AparteChatMessage, AparteStreamEvent, AparteUsage } from '@aparte/core';
+import type { BuiltInRunner, Device, Dtype, GenerationOptions, RunnerModule, RunnerProgress, TransformersModule } from './runners/types.js';
+interface RunnerSelection {
+    modelId: string;
+    task?: BuiltInRunner;
+    runner?: string;
+    dtype?: Dtype;
+    device?: Device;
+}
+export type InMessage = {
+    type: 'init';
+    transformersUrl?: string;
+} | ({
+    type: 'prepare';
+    id: string;
+} & RunnerSelection) | ({
+    type: 'generate';
+    id: string;
+    messages: AparteChatMessage[];
+    options: GenerationOptions;
+} & RunnerSelection) | {
+    type: 'cancel';
+    id: string;
+} | ({
+    type: 'command';
+    id: string;
+    name: string;
+    payload: unknown;
+} & RunnerSelection);
+export type OutMessage = ({
+    type: 'progress';
+    id: string;
+} & RunnerProgress) | {
+    type: 'pipeline-ready';
+    modelId: string;
+} | {
+    type: 'prepare-error';
+    id: string;
+    message: string;
+} | {
+    type: 'gen-event';
+    id: string;
+    event: AparteStreamEvent;
+} | {
+    type: 'gen-done';
+    id: string;
+    usage?: AparteUsage;
+} | {
+    type: 'gen-error';
+    id: string;
+    message: string;
+} | {
+    type: 'warning';
+    message: string;
+} | {
+    type: 'command-result';
+    id: string;
+    result?: unknown;
+    error?: string;
+};
+export interface WorkerHostDeps {
+    post(message: OutMessage): void;
+    /** Resolve Transformers.js — the bundled specifier first, else the URL `init` carried. */
+    loadTransformers(moduleUrl?: string): Promise<TransformersModule>;
+    /** Import the runner module a selection names: a built-in by `task`, a custom one by `runner` URL. */
+    importRunner(spec: {
+        task: BuiltInRunner;
+        runner?: string;
+    }): Promise<RunnerModule>;
+}
+export declare function createWorkerHost(deps: WorkerHostDeps): {
+    onMessage(msg: InMessage): void;
+};
+export {};
+//# sourceMappingURL=worker-host.d.ts.map

package/dist/worker-host.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"worker-host.d.ts","sourceRoot":"","sources":["../src/worker-host.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,OAAO,KAAK,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AACtF,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,EAAE,KAAK,EAAE,iBAAiB,EAAE,YAAY,EAAE,cAAc,EAAE,kBAAkB,EAAsB,MAAM,oBAAoB,CAAC;AAEhK,UAAU,eAAe;IACrB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,MAAM,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,MAAM,SAAS,GACf;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,eAAe,CAAC,EAAE,MAAM,CAAA;CAAE,GAC1C,CAAC;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAAG,eAAe,CAAC,GACnD,CAAC;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAAC,OAAO,EAAE,iBAAiB,CAAA;CAAE,GAAG,eAAe,CAAC,GAC/G;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAC9B,CAAC;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,OAAO,CAAA;CAAE,GAAG,eAAe,CAAC,CAAC;AAE1F,MAAM,MAAM,UAAU,GAChB,CAAC;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAAG,cAAc,CAAC,GACnD;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC3C;IAAE,IAAI,EAAE,eAAe,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACtD;IAAE,IAAI,EAAE,WAAW,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,iBAAiB,CAAA;CAAE,GAC3D;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,KAAK,CAAC,EAAE,WAAW,CAAA;CAAE,GACrD;IAAE,IAAI,EAAE,WAAW,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAClD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GACpC;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC;AAE/E,MAAM,WAAW,cAAc;IAC3B,IAAI,CAAC,OAAO,EAAE,UAAU,GAAG,IAAI,CAAC;IAChC,0FAA0F;IAC1F,gBAAgB,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAAC;IAClE,sGAAsG;IACtG,YAAY,CAAC,IAAI,EAAE;QAAE,IAAI,EAAE,aAAa,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;CACvF;AAID,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,cAAc,GAAG;IAAE,SAAS,CAAC,GAAG,EAAE,SAAS,GAAG,IAAI,CAAA;CAAE,CAkG1F"}

package/dist/worker.d.ts

@@ -1,21 +1,10 @@
 /**
- * Generic Transformers.js inference worker.
+ * The Transformers.js inference worker — the shell.
  *
- * Runs entirely off the main thread. It holds ONE text-generation pipeline at a
- * time and speaks a tiny postMessage protocol with the provider on the main
- * thread (see `index.ts`):
- *
- *   main → worker : { type: 'prepare',  id, modelId, dtype?, device? }
- *                   { type: 'generate', id, modelId, messages, options, dtype?, device? }
- *   worker → main : { type: 'progress', id, status, file?, progress? }
- *                   { type: 'prepare-error', id, message }
- *                   { type: 'pipeline-ready', modelId }
- *                   { type: 'gen-chunk', id, chunkType: 'text', delta }
- *                   { type: 'gen-done', id }
- *                   { type: 'gen-error', id, message }
- *
- * Deliberately generic: no vision, no low-level ORT session management, no
- * model-family specifics — just the high-level `pipeline()` + `TextStreamer`.
+ * Runs entirely off the main thread. What it decides lives in `worker-host.ts` (the
+ * protocol, the one-runner-at-a-time rule, cancel, warnings); what a model IS lives in a
+ * runner (`runners/`). This file binds the two to `self` and owns the two things only a
+ * real worker can do: resolve Transformers.js, and import a runner module.
  */
 export {};
 //# sourceMappingURL=worker.d.ts.map

package/dist/worker.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"worker.d.ts","sourceRoot":"","sources":["../src/worker.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG"}
+{"version":3,"file":"worker.d.ts","sourceRoot":"","sources":["../src/worker.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aparte/provider-transformers",
-  "version": "0.16.3",
+  "version": "0.16.5",
   "description": "Run LLMs 100% in the browser via Transformers.js (WebGPU/WASM) — a local, keyless AI provider for aparté. Streams tokens off the main thread in a Web Worker.",
   "type": "module",
   "sideEffects": false,
@@ -24,7 +24,7 @@
     "node": ">=18"
   },
   "peerDependencies": {
-    "@aparte/core": ">=0.16.3 <1.0.0",
+    "@aparte/core": ">=0.16.5 <1.0.0",
     "@huggingface/transformers": "^4.2.0"
   },
   "devDependencies": {
@@ -32,7 +32,7 @@
     "@types/node": "^22.0.0",
     "typescript": "^5.4.0",
     "vite": "^6.0.0",
-    "@aparte/core": "0.16.3"
+    "@aparte/core": "0.16.5"
   },
   "keywords": [
     "ai",

package/dist/assets/worker-Dz5bU2S7.js

@@ -1,93 +0,0 @@
-let _moduleUrl;
-let _tf = null;
-function transformers() {
-  _tf ??= (async () => {
-    let mod;
-    try {
-      mod = await import("@huggingface/transformers");
-    } catch (bundlerPathFailed) {
-      if (!_moduleUrl) throw bundlerPathFailed;
-      mod = await import(
-        /* @vite-ignore */
-        _moduleUrl
-      );
-    }
-    mod.env.allowLocalModels = false;
-    mod.env.useBrowserCache = true;
-    return mod;
-  })();
-  return _tf;
-}
-const ctx = self;
-function post(message) {
-  ctx.postMessage(message);
-}
-let _current = null;
-const _activeStops = /* @__PURE__ */ new Map();
-async function ensurePipeline(modelId, dtype, device, id) {
-  if (_current?.modelId === modelId) return _current.pipe;
-  const opts = {
-    progress_callback: (p) => {
-      if (!id) return;
-      if (p.status === "progress") {
-        post({ type: "progress", id, status: "downloading", file: p.file, progress: Math.round(p.progress ?? 0) });
-      } else if (p.status === "done") {
-        post({ type: "progress", id, status: "loading", file: p.file });
-      }
-    }
-  };
-  if (dtype) opts["dtype"] = dtype;
-  if (device && device !== "auto") opts["device"] = device;
-  const { pipeline } = await transformers();
-  const pipe = await pipeline("text-generation", modelId, opts);
-  _current = { modelId, pipe };
-  post({ type: "pipeline-ready", modelId });
-  return pipe;
-}
-async function handlePrepare(msg) {
-  try {
-    await ensurePipeline(msg.modelId, msg.dtype, msg.device, msg.id);
-    post({ type: "progress", id: msg.id, status: "ready" });
-  } catch (err) {
-    post({ type: "prepare-error", id: msg.id, message: err?.message ?? "Failed to load model" });
-  }
-}
-async function handleGenerate(msg) {
-  const { TextStreamer, InterruptableStoppingCriteria } = await transformers();
-  const stoppingCriteria = new InterruptableStoppingCriteria();
-  _activeStops.set(msg.id, stoppingCriteria);
-  try {
-    const pipe = await ensurePipeline(msg.modelId, msg.dtype, msg.device, msg.id);
-    const streamer = new TextStreamer(pipe.tokenizer, {
-      skip_prompt: true,
-      skip_special_tokens: true,
-      callback_function: (text) => {
-        if (text) post({ type: "gen-chunk", id: msg.id, chunkType: "text", delta: text });
-      }
-    });
-    const temperature = msg.options.temperature ?? 0;
-    await pipe(msg.messages, {
-      max_new_tokens: msg.options.maxTokens ?? 512,
-      do_sample: temperature > 0,
-      temperature: temperature > 0 ? temperature : void 0,
-      streamer,
-      stopping_criteria: stoppingCriteria
-    });
-    post({ type: "gen-done", id: msg.id });
-  } catch (err) {
-    post({ type: "gen-error", id: msg.id, message: err?.message ?? "Generation failed" });
-  } finally {
-    _activeStops.delete(msg.id);
-  }
-}
-ctx.addEventListener("message", (event) => {
-  const msg = event.data;
-  if (msg.type === "init") {
-    _moduleUrl = msg.transformersUrl;
-    return;
-  }
-  if (msg.type === "prepare") void handlePrepare(msg);
-  else if (msg.type === "generate") void handleGenerate(msg);
-  else if (msg.type === "cancel") _activeStops.get(msg.id)?.interrupt();
-});
-//# sourceMappingURL=worker-Dz5bU2S7.js.map

package/dist/assets/worker-Dz5bU2S7.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"worker-Dz5bU2S7.js","sources":["../src/worker.ts"],"sourcesContent":["/**\n * Generic Transformers.js inference worker.\n *\n * Runs entirely off the main thread. It holds ONE text-generation pipeline at a\n * time and speaks a tiny postMessage protocol with the provider on the main\n * thread (see `index.ts`):\n *\n *   main → worker : { type: 'prepare',  id, modelId, dtype?, device? }\n *                   { type: 'generate', id, modelId, messages, options, dtype?, device? }\n *   worker → main : { type: 'progress', id, status, file?, progress? }\n *                   { type: 'prepare-error', id, message }\n *                   { type: 'pipeline-ready', modelId }\n *                   { type: 'gen-chunk', id, chunkType: 'text', delta }\n *                   { type: 'gen-done', id }\n *                   { type: 'gen-error', id, message }\n *\n * Deliberately generic: no vision, no low-level ORT session management, no\n * model-family specifics — just the high-level `pipeline()` + `TextStreamer`.\n */\n\nimport type { pipeline as Pipeline, TextStreamer as TextStreamerClass, InterruptableStoppingCriteria as StoppingCriteriaClass, TextGenerationPipeline } from '@huggingface/transformers';\n\n/**\n * Where Transformers.js comes from, resolved once, from whichever path has it.\n *\n * A static `import … from '@huggingface/transformers'` is unresolvable in a worker\n * served without a bundler: an import map lives on the DOCUMENT and, by spec, does not\n * reach a worker — so the page can map the specifier for itself and the worker still\n * cannot. The two paths, in order:\n *\n *   1. `import('@huggingface/transformers')` — a bare specifier, statically visible, so a\n *      consumer's bundler resolves and bundles the peer exactly as it did before.\n *   2. the absolute URL the main thread read from the page's own import map and sent in\n *      the first message — the CDN path, where that map is the consumer's manifest.\n *\n * The order matters: a bundled app must never reach for the network copy.\n */\ntype TransformersModule = {\n    pipeline: typeof Pipeline;\n    TextStreamer: typeof TextStreamerClass;\n    InterruptableStoppingCriteria: typeof StoppingCriteriaClass;\n    env: { allowLocalModels: boolean; useBrowserCache: boolean };\n};\n\nlet _moduleUrl: string | undefined;\nlet _tf: Promise<TransformersModule> | null = null;\n\nfunction transformers(): Promise<TransformersModule> {\n    _tf ??= (async () => {\n        let mod: TransformersModule;\n        try {\n            mod = await import('@huggingface/transformers') as unknown as TransformersModule;\n        } catch (bundlerPathFailed) {\n            if (!_moduleUrl) throw bundlerPathFailed;\n            mod = await import(/* @vite-ignore */ _moduleUrl) as unknown as TransformersModule;\n        }\n        // Fetch weights from the Hugging Face hub (not local paths) and cache them in the\n        // browser Cache API — this is what `listCachedModels()` scans on the main thread.\n        mod.env.allowLocalModels = false;\n        mod.env.useBrowserCache = true;\n        return mod;\n    })();\n    return _tf;\n}\n\n// DOM's `Worker` interface types `postMessage` + typed `addEventListener('message')`,\n// which is enough for the worker scope — avoids pulling the WebWorker lib (it clashes\n// with DOM's global `postMessage`).\nconst ctx = self as unknown as Worker;\n\ntype Dtype = string | Record<string, string>;\ntype Device = 'webgpu' | 'wasm' | 'auto';\ninterface GenOptions { maxTokens?: number; temperature?: number; seed?: number }\ntype SimpleMessage = { role: 'user' | 'assistant' | 'system'; content: string };\n\ntype InMessage =\n    | { type: 'prepare'; id: string; modelId: string; dtype?: Dtype; device?: Device }\n    | { type: 'generate'; id: string; modelId: string; messages: SimpleMessage[]; options: GenOptions; dtype?: Dtype; device?: Device }\n    | { type: 'cancel'; id: string }\n    // Sent once, before anything else, when the main thread could read the peer's URL out\n    // of the page's import map. Absent under a bundler, which has already resolved it.\n    | { type: 'init'; transformersUrl?: string };\n\nfunction post(message: unknown): void {\n    ctx.postMessage(message);\n}\n\nlet _current: { modelId: string; pipe: TextGenerationPipeline } | null = null;\n// Per-generate interrupts, so a consumer's stream-cancel actually STOPS the model\n// (not just detaches the reader) — otherwise generation runs to max_new_tokens\n// off-thread, wasting exactly the CPU/GPU/battery this provider exists to save.\nconst _activeStops = new Map<string, InstanceType<typeof StoppingCriteriaClass>>();\n\n/**\n * Ensure the pipeline for `modelId` is loaded, reusing the current one when it\n * matches. On a fresh load it forwards download progress (when `id` is given) and\n * announces `pipeline-ready`.\n */\nasync function ensurePipeline(modelId: string, dtype: Dtype | undefined, device: Device | undefined, id?: string): Promise<TextGenerationPipeline> {\n    if (_current?.modelId === modelId) return _current.pipe;\n\n    // eslint-disable-next-line @typescript-eslint/no-explicit-any\n    const opts: Record<string, any> = {\n        progress_callback: (p: { status?: string; file?: string; progress?: number }) => {\n            if (!id) return;\n            if (p.status === 'progress') {\n                post({ type: 'progress', id, status: 'downloading', file: p.file, progress: Math.round(p.progress ?? 0) });\n            } else if (p.status === 'done') {\n                post({ type: 'progress', id, status: 'loading', file: p.file });\n            }\n        },\n    };\n    if (dtype) opts['dtype'] = dtype;\n    if (device && device !== 'auto') opts['device'] = device;\n\n    const { pipeline } = await transformers();\n    const pipe = await pipeline('text-generation', modelId, opts) as TextGenerationPipeline;\n    _current = { modelId, pipe };\n    post({ type: 'pipeline-ready', modelId });\n    return pipe;\n}\n\nasync function handlePrepare(msg: Extract<InMessage, { type: 'prepare' }>): Promise<void> {\n    try {\n        await ensurePipeline(msg.modelId, msg.dtype, msg.device, msg.id);\n        post({ type: 'progress', id: msg.id, status: 'ready' });\n    } catch (err) {\n        post({ type: 'prepare-error', id: msg.id, message: (err as Error)?.message ?? 'Failed to load model' });\n    }\n}\n\nasync function handleGenerate(msg: Extract<InMessage, { type: 'generate' }>): Promise<void> {\n    const { TextStreamer, InterruptableStoppingCriteria } = await transformers();\n    const stoppingCriteria = new InterruptableStoppingCriteria();\n    _activeStops.set(msg.id, stoppingCriteria);\n    try {\n        const pipe = await ensurePipeline(msg.modelId, msg.dtype, msg.device, msg.id);\n\n        const streamer = new TextStreamer(pipe.tokenizer, {\n            skip_prompt: true,\n            skip_special_tokens: true,\n            callback_function: (text: string) => {\n                if (text) post({ type: 'gen-chunk', id: msg.id, chunkType: 'text', delta: text });\n            },\n        });\n\n        const temperature = msg.options.temperature ?? 0;\n        await pipe(msg.messages, {\n            max_new_tokens: msg.options.maxTokens ?? 512,\n            do_sample: temperature > 0,\n            temperature: temperature > 0 ? temperature : undefined,\n            streamer,\n            stopping_criteria: stoppingCriteria,\n        });\n\n        post({ type: 'gen-done', id: msg.id });\n    } catch (err) {\n        post({ type: 'gen-error', id: msg.id, message: (err as Error)?.message ?? 'Generation failed' });\n    } finally {\n        _activeStops.delete(msg.id);\n    }\n}\n\nctx.addEventListener('message', (event: MessageEvent<InMessage>) => {\n    const msg = event.data;\n    if (msg.type === 'init') { _moduleUrl = msg.transformersUrl; return; }\n    if (msg.type === 'prepare') void handlePrepare(msg);\n    else if (msg.type === 'generate') void handleGenerate(msg);\n    else if (msg.type === 'cancel') _activeStops.get(msg.id)?.interrupt();\n});\n"],"names":[],"mappings":"AA4CA,IAAI;AACJ,IAAI,MAA0C;AAE9C,SAAS,eAA4C;AACjD,WAAS,YAAY;AACjB,QAAI;AACJ,QAAI;AACA,YAAM,MAAM,OAAO,2BAA2B;AAAA,IAClD,SAAS,mBAAmB;AACxB,UAAI,CAAC,WAAY,OAAM;AACvB,YAAM,MAAM;AAAA;AAAA,QAA0B;AAAA;AAAA,IAC1C;AAGA,QAAI,IAAI,mBAAmB;AAC3B,QAAI,IAAI,kBAAkB;AAC1B,WAAO;AAAA,EACX,GAAA;AACA,SAAO;AACX;AAKA,MAAM,MAAM;AAeZ,SAAS,KAAK,SAAwB;AAClC,MAAI,YAAY,OAAO;AAC3B;AAEA,IAAI,WAAqE;AAIzE,MAAM,mCAAmB,IAAA;AAOzB,eAAe,eAAe,SAAiB,OAA0B,QAA4B,IAA8C;AAC/I,MAAI,UAAU,YAAY,QAAS,QAAO,SAAS;AAGnD,QAAM,OAA4B;AAAA,IAC9B,mBAAmB,CAAC,MAA6D;AAC7E,UAAI,CAAC,GAAI;AACT,UAAI,EAAE,WAAW,YAAY;AACzB,aAAK,EAAE,MAAM,YAAY,IAAI,QAAQ,eAAe,MAAM,EAAE,MAAM,UAAU,KAAK,MAAM,EAAE,YAAY,CAAC,GAAG;AAAA,MAC7G,WAAW,EAAE,WAAW,QAAQ;AAC5B,aAAK,EAAE,MAAM,YAAY,IAAI,QAAQ,WAAW,MAAM,EAAE,MAAM;AAAA,MAClE;AAAA,IACJ;AAAA,EAAA;AAEJ,MAAI,MAAO,MAAK,OAAO,IAAI;AAC3B,MAAI,UAAU,WAAW,OAAQ,MAAK,QAAQ,IAAI;AAElD,QAAM,EAAE,aAAa,MAAM,aAAA;AAC3B,QAAM,OAAO,MAAM,SAAS,mBAAmB,SAAS,IAAI;AAC5D,aAAW,EAAE,SAAS,KAAA;AACtB,OAAK,EAAE,MAAM,kBAAkB,QAAA,CAAS;AACxC,SAAO;AACX;AAEA,eAAe,cAAc,KAA6D;AACtF,MAAI;AACA,UAAM,eAAe,IAAI,SAAS,IAAI,OAAO,IAAI,QAAQ,IAAI,EAAE;AAC/D,SAAK,EAAE,MAAM,YAAY,IAAI,IAAI,IAAI,QAAQ,SAAS;AAAA,EAC1D,SAAS,KAAK;AACV,SAAK,EAAE,MAAM,iBAAiB,IAAI,IAAI,IAAI,SAAU,KAAe,WAAW,uBAAA,CAAwB;AAAA,EAC1G;AACJ;AAEA,eAAe,eAAe,KAA8D;AACxF,QAAM,EAAE,cAAc,8BAAA,IAAkC,MAAM,aAAA;AAC9D,QAAM,mBAAmB,IAAI,8BAAA;AAC7B,eAAa,IAAI,IAAI,IAAI,gBAAgB;AACzC,MAAI;AACA,UAAM,OAAO,MAAM,eAAe,IAAI,SAAS,IAAI,OAAO,IAAI,QAAQ,IAAI,EAAE;AAE5E,UAAM,WAAW,IAAI,aAAa,KAAK,WAAW;AAAA,MAC9C,aAAa;AAAA,MACb,qBAAqB;AAAA,MACrB,mBAAmB,CAAC,SAAiB;AACjC,YAAI,KAAM,MAAK,EAAE,MAAM,aAAa,IAAI,IAAI,IAAI,WAAW,QAAQ,OAAO,KAAA,CAAM;AAAA,MACpF;AAAA,IAAA,CACH;AAED,UAAM,cAAc,IAAI,QAAQ,eAAe;AAC/C,UAAM,KAAK,IAAI,UAAU;AAAA,MACrB,gBAAgB,IAAI,QAAQ,aAAa;AAAA,MACzC,WAAW,cAAc;AAAA,MACzB,aAAa,cAAc,IAAI,cAAc;AAAA,MAC7C;AAAA,MACA,mBAAmB;AAAA,IAAA,CACtB;AAED,SAAK,EAAE,MAAM,YAAY,IAAI,IAAI,IAAI;AAAA,EACzC,SAAS,KAAK;AACV,SAAK,EAAE,MAAM,aAAa,IAAI,IAAI,IAAI,SAAU,KAAe,WAAW,oBAAA,CAAqB;AAAA,EACnG,UAAA;AACI,iBAAa,OAAO,IAAI,EAAE;AAAA,EAC9B;AACJ;AAEA,IAAI,iBAAiB,WAAW,CAAC,UAAmC;AAChE,QAAM,MAAM,MAAM;AAClB,MAAI,IAAI,SAAS,QAAQ;AAAE,iBAAa,IAAI;AAAiB;AAAA,EAAQ;AACrE,MAAI,IAAI,SAAS,UAAW,MAAK,cAAc,GAAG;AAAA,WACzC,IAAI,SAAS,WAAY,MAAK,eAAe,GAAG;AAAA,WAChD,IAAI,SAAS,SAAU,cAAa,IAAI,IAAI,EAAE,GAAG,UAAA;AAC9D,CAAC;"}