@qvac/tts-ggml 0.11.1 → 0.13.0

package/README.md

@@ -29,7 +29,8 @@ Xclipse/Mali → Vulkan). Parler supports Apple/Metal, linux CUDA, and the
 validated Android paths, including Vulkan on ARM Mali (see
 [Backends & GPU acceleration](#backends--gpu-acceleration)). Audio8 supports
 CUDA/Vulkan offload on Linux and Vulkan on Windows. MOSS GPU offload is
-validated on Apple/Metal.
+validated on Apple/Metal; desktop CUDA/Vulkan validation requires the native
+and registry prerequisites described under [MOSS](#moss).
 
 [qvac-tts-cpp]: https://github.com/tetherto/qvac-fabric-speech.cpp/tree/master/engines/tts
 
@@ -54,8 +55,9 @@ validated on Apple/Metal.
   iOS, picked up when staged next to the model (see
   [Core ML sidecars on Apple](#core-ml-sidecars-on-apple)).
 - **Dynamic backend loading on Android** — per-arch CPU + Vulkan +
-  OpenCL `.so` files ship under `prebuilds/<bare-target>/qvac__tts-ggml/`
-  and are picked up at runtime via the new `backendsDir` option (see
+  OpenCL `.so` files ship under
+  `prebuilds/<bare-target>/qvac__tts-ggml-android-arm64/` and the addon picks
+  them up at runtime from next to its native module (see
   [Backends & GPU acceleration](#backends--gpu-acceleration)).
 - **Cancellation** via `model.cancel()` — stops T3 decode on the next
   token; in-flight S3Gen chunk runs to completion.
@@ -150,11 +152,10 @@ package selected at install time through `os`/`cpu` filtered
 
 Do not depend on desktop platform packages directly. Supported installers are
 npm 7+, pnpm, bun, and Yarn Berry. Yarn v1 and `--omit=optional` installs skip
-the platform package and fail at require time with an error naming the missing
-package; a locally built `prebuilds/` directory in the package root always
-takes precedence. Use `require('@qvac/tts-ggml').resolveBackendsDir()` to
-locate the directory holding the host's prebuilt binaries and dynamically
-loaded ggml backends. Unsupported targets must
+the platform package and fail at require time, and at `bare-pack` time, naming
+the missing package. Each platform package is an ordinary Bare addon whose
+`prebuilds/<host>/` holds the native module and, next to it, the dynamically
+loaded ggml backends the module finds on its own. Unsupported targets must
 [build from source](#build-from-source); installation does not automatically
 compile a local addon.
 
@@ -805,13 +806,13 @@ registration and the addon falls back to Vulkan or CPU.
 
 Android prebuilds enable `GGML_BACKEND_DL=ON` and ship per-arch
 backend `.so` files under
-`prebuilds/<bare-target>/qvac__tts-ggml/`.
+`prebuilds/android-arm64/qvac__tts-ggml-android-arm64/` of
+`@qvac/tts-ggml-android-arm64`; `bare-link` puts them next to the module in
+the app, where the addon looks when `backendsDir` is unset.
 
 The engine `dlopen()`s the highest-tier CPU variant the device's
 HWCAPs support and one of the GPU `.so` files based on the policy
-table above.  Hosts must pass `backendsDir: path.join(__dirname,
-'prebuilds')` (or rely on the default fallback the package ships)
-so the runtime knows where to look.  `openclCacheDir` is also
+table above.  `openclCacheDir` is
 Android-specific; setting it to a writable path lets the OpenCL
 backend persist its compiled program cache across launches.
 `vulkanCacheDir` is the Vulkan analogue (Supertonic + `useGPU: true`):
@@ -1060,6 +1061,82 @@ falls back to CPU and sets `response.stats.gpuUnsupported`.  `backendsDir`
 reaches MOSS like every other engine, so builds that load ggml backends at
 runtime find them.
 
+The same controls apply to MOSS-TTSD and MOSS-Speech. `nGpuLayers` is a
+boolean GPU request for these engines (zero = CPU, non-zero = GPU), not a
+partial layer count. The addon forwards these controls to the native engine;
+it does not choose a different backend or change the CPU default.
+`response.stats.backendId` identifies the language-model backend (2 = CUDA,
+3 = Vulkan); it is not a measurement of every codec operation.
+
+The combined desktop CUDA/Vulkan speech changes in
+[Vulkan #308](https://github.com/tetherto/qvac-fabric-speech.cpp/pull/308),
+including [CUDA #321](https://github.com/tetherto/qvac-fabric-speech.cpp/pull/321),
+merged to native master at `84b02694001cfea863167776dbc185474a816656`.
+This checkout consumes the merged [registry #423](https://github.com/tetherto/qvac-registry-vcpkg/pull/423)
+at `e31be272ecdf9f20970302ebf8a2943d47803001`: `speech-cpp` 2026-10-12#1
+pins that merged revision, and `ggml-speech` 2026-10-07#1 pins
+`06f1b66806626a635904f31db6914ad2d0e71197`. The earlier native `f4d4c257`
+passed ten smaller-model CUDA/Vulkan cases; those results do not establish GPU
+validation of the final native revision or the packaged addon. Rebuild the addon
+before running its GPU tests. Native desktop results also do not establish
+Android support or speech-quality signoff.
+
+The registry `baseline` is the main commit containing #423. The shared exact
+`speech-cpp` overrides and backend implementation are already on qvac main.
+This baseline update completes the TTS registry configuration; packaged-addon
+CUDA/Vulkan inference still requires validation on provisioned GPU hosts before
+claiming a validated release.
+
+From `packages/tts-ggml`, resolve the checked-in manifest without building or
+downloading models (with `VCPKG_ROOT` set):
+
+```sh
+vcpkg install --dry-run --triplet x64-linux --x-feature=cuda --x-feature=vulkan
+```
+
+On a macOS host without a Linux cross-compiler, prefix that dry-run command
+with `VCPKG_FEATURE_FLAGS=-compilertracking` to check registry resolution only.
+For a bounded native build on a suitably provisioned Linux GPU host:
+
+```sh
+CMAKE_BUILD_PARALLEL_LEVEL=2 VCPKG_MAX_CONCURRENCY=2 npm run build:cuda
+```
+
+The normal PR workflow requires approval of the `fork-ci` environment for
+external forks. Obtain that approval through the existing gate; a manual
+workflow dispatch is not a substitute for fork approval.
+
+On a desktop with those prerequisites, stage the TTS and TTSD backbones,
+codec decoder/encoder and the bundled `test/reference-audio/jfk.wav`, plus
+the Speech model/codec. Set
+`QVAC_TEST_MOSS_MODEL_DIR` and `QVAC_TEST_MOSS_SPEECH_MODEL_DIR` to their
+respective directories. Run each available backend in a fresh process:
+
+```sh
+QVAC_TEST_MOSS_FULL_GPU=true TTS_CPP_GPU_BACKEND=cuda npx --no-install brittle-bare test/integration/moss.test.js test/integration/moss-speech.test.js
+QVAC_TEST_MOSS_FULL_GPU=true TTS_CPP_GPU_BACKEND=vulkan npx --no-install brittle-bare test/integration/moss.test.js test/integration/moss-speech.test.js
+```
+
+The tests check the exact pinned backend, `backendDevice` and
+`gpuUnsupported` for TTS batch/streaming/cloning, TTSD dialogue, and Speech
+text/audio replies. CPU control cases still require CPU even with a GPU pin.
+`QVAC_TEST_MOSS_FULL_GPU=true` fails before model loading unless both model
+directories, the TTS/TTSD/Speech backbones, all three codec GGUFs, and the
+reference WAV exist. It also requires an explicit backend pin (case-insensitive),
+rejects `NO_GPU=true` and mobile CPU-only cases, and checks that every required
+case completes. Run both files together without test filters: a successful full
+run has 12 passing tests, including the two completion checks, and no skips.
+Preflight only checks asset presence; actual model loading and inference must
+also pass. These commands cover TTS, TTSD and Speech, not MOSS-SoundEffect (SFX).
+No addon CUDA/Vulkan hardware run has been completed for this test change yet.
+
+Without strict mode, developer behavior is unchanged: an unpinned run accepts
+the normal GPU cascade, missing model directories skip tests, and missing
+encoder/TTSD assets omit cloning/dialogue cases. Such runs are not full GPU
+validation. `npm run test:node` covers backend selection and strict validation
+failures without loading models. Do not load these multi-GB models on
+memory-constrained hosts.
+
 ## MOSS-SoundEffect
 
 MOSS-SoundEffect-v2 generates a sound effect from a text description: a Qwen3
@@ -1228,7 +1305,7 @@ instance.  Runtime stats add `promptTokens`, `generatedTokens`, `replyTokens`,
 | `promptText`              | string     | —          | CosyVoice3-only: verbatim transcript of `referenceAudio` — set for zero-shot cloning, omit for cross-lingual; without a reference it overrides the baked voice's transcript |
 | `mecabDictDir`            | string     | —          | Chatterbox MTL Japanese (`ja`): compiled MeCab/IPAdic dictionary directory |
 | `cangjieTsvPath`          | string     | —          | Chatterbox MTL Chinese (`zh`): `Cangjie5_TC` TSV path |
-| `backendsDir`             | string     | `resolveBackendsDir()` | Root dir the addon scans for dynamically-loaded ggml backend `.so` files.  Defaults to the package's own `prebuilds/` when present, otherwise the installed platform package.  Required on Android when backends ship elsewhere (e.g. inside the APK); ignored on platforms that statically link the backend |
+| `backendsDir`             | string     | the module's own | Directory the addon scans, as given, for dynamically-loaded ggml backend `.so` / `.dll` files.  Unset scans the directory the native module was loaded from (`prebuilds/<host>/qvac__tts-ggml-<suffix>/` in a platform package, the app's native library directory after `bare-link`); ignored on platforms that statically link the backend |
 | `openclCacheDir`          | string     | unset      | Android-only: directory where the OpenCL backend persists its compiled program-binary cache.  Setting it across runs avoids re-JITing the kernels on every fresh process |
 | `vulkanCacheDir`          | string     | unset      | Supertonic + `useGPU: true` only: writable directory where the Vulkan backend persists its compiled pipeline cache (`GGML_VK_PIPELINE_CACHE_DIR`).  Moves the one-time first-dispatch pipeline-compile cost (seconds on Mali) off the first `run()` — paid once per install instead of once per process — and enables a load-time pre-warm.  Fully opt-in: unset -> no cross-process cache, no pre-warm, behaviour unchanged |
 | `config.language`         | string     | `"en"`     | Chatterbox MTL accepts `es/fr/de/pt/it/zh/ja/ko/...`; turbo & Supertonic are English; MOSS takes it as a prompt hint (`en`, `zh`, ...) |
@@ -1348,12 +1425,14 @@ const {
 
 `assessFit` projects a load against the memory free right now. It reads GGUF metadata and never weight data, so the registry's weightless copy of each file answers the same as the file itself and the projection can run before anything is downloaded. It is a module export, not an instance method: nothing is loaded to call it.
 
+`assessFit` returns a Promise. Fits run one at a time within this addon.
+
 A request is a `createInstance` config plus the workload, so it carries the same keys the load takes and the engine's own config builder reads it.
 
 ```js
 const TTSGgml = require('@qvac/tts-ggml')
 
-const fit = TTSGgml.assessFit({
+const fit = await TTSGgml.assessFit({
   engineType: 'supertonic',
   supertonicModelPath: '/models/supertonic.gguf',
   useGPU: true,
@@ -1393,7 +1472,7 @@ Everything else comes from the load config: `nGpuLayers` and `useGPU` carry the
 
 `deviceSharesHostMemory` reports that the device pool is system RAM, so host bytes compete with device bytes.
 
-A model the engine cannot read is `status: "error"`, as is a voice with no fitter, which reports `reason: "unsupported-engine"`. MOSS-TTS, MOSS-TTSD and MOSS-SoundEffect are supported. MOSS-Speech has no SDK fit projection. A broken request, or a host with no native binding, throws.
+A model the engine cannot read resolves to `status: "error"`, as does a voice with no fitter, which reports `reason: "unsupported-engine"`. MOSS-TTS, MOSS-TTSD and MOSS-SoundEffect are supported. MOSS-Speech has no SDK fit projection. A broken request, or a host with no native binding, rejects.
 
 MOSS-TTS and MOSS-TTSD share the `moss` fitter. `promptRows` is the complete
 native prompt length, including special tokens, encoded speaker references and
@@ -1404,7 +1483,7 @@ The fitter does not open reference recordings. Use the TTSD backbone to assess
 dialogue, supplying its complete prompt and total reference workload.
 
 ```js
-const fit = TTSGgml.assessFit({
+const fit = await TTSGgml.assessFit({
   engineType: 'moss',
   mossBackbonePath: './moss-tts-delay-f16.gguf',
   mossCodecDecoderPath: './moss-codec-decoder-f16.gguf',
@@ -1426,7 +1505,7 @@ The supertonic fitter covers the fused graph path: a validated GPU, or a CPU wit
 ### MOSS-SoundEffect fit
 
 ```js
-const fit = TTSGgml.assessFit({
+const fit = await TTSGgml.assessFit({
   engineType: 'moss-sfx',
   mossSoundEffectPath: './moss-sfx-v2-q8_0.gguf',
   prompt: 'Rain falling on a tin roof.',
@@ -1571,11 +1650,18 @@ npm install
 npx bare-make generate      # configures + fetches the tts-cpp port
 npx bare-make build
 npx bare-make install       # copies the .bare into prebuilds/<triple>/
+npm run link:platform       # stages node_modules/@qvac/tts-ggml-<host>
 ```
 
+`require('#host-addon')` only ever loads a platform package, so a source
+build becomes loadable once `npm run link:platform` (run by
+`npm run build:native`) wraps `prebuilds/<host>/` as that host's
+`@qvac/tts-ggml-<host>` package in `node_modules/`.
+
 The vcpkg port is hosted in
 [`tetherto/qvac-registry-vcpkg`][registry] and pulls
-[`qvac-tts.cpp`][qvac-tts-cpp] at a pinned REF.  See
+[`qvac-fabric-speech.cpp`](https://github.com/tetherto/qvac-fabric-speech.cpp)
+(`engines/tts`) at a pinned REF.  See
 [`vcpkg-configuration.json`](./vcpkg-configuration.json) for the
 baseline commit.
 

package/addon-unavailable.js

@@ -1,31 +1,15 @@
-const { hostPlatformPackage, PREBUILT_HOSTS } = require('./lib/backends.js')
+'use strict'
 
-const META_PACKAGE = '@qvac/tts-ggml'
-const UNKNOWN_HOST = 'unknown'
+// "#host-addon" target for hosts with no platform package. Hosts that have one
+// map straight to it with no fallback, so a missing install fails bare-pack (and
+// fails at run time naming the package) instead of producing a bundle that only
+// throws once launched.
+const host = require.addon ? require.addon.host : 'unknown'
 
-throw new Error(buildMessage(currentHost()))
-
-function currentHost() {
-  return require.addon ? require.addon.host : null
-}
-
-function buildMessage(host) {
-  if (!host || !PREBUILT_HOSTS.includes(host)) {
-    return (
-      META_PACKAGE +
-      ' has no prebuilt binaries for host ' +
-      (host || UNKNOWN_HOST) +
-      '. Prebuilt hosts: ' +
-      PREBUILT_HOSTS.join(', ') +
-      '. Build from source with bare-make.'
-    )
-  }
-  return (
-    META_PACKAGE +
-    ' found no native prebuild for ' +
+throw new Error(
+  '@qvac/tts-ggml has no prebuilt binaries for host ' +
     host +
-    ': the platform package ' +
-    hostPlatformPackage(host) +
-    ' is not installed. It ships as an os/cpu filtered optional dependency, which Yarn v1 and installs using --omit=optional drop. Reinstall with npm 7+, pnpm, bun, or Yarn Berry, or build from source with bare-make.'
-  )
-}
+    '. Platform packages exist for linux-x64, linux-arm64, darwin-arm64, ' +
+    'darwin-x64, win32-x64, android-arm64 and ios (ios-arm64, ' +
+    'ios-arm64-simulator, ios-x64-simulator). Build from source with bare-make.'
+)

package/binding.js

@@ -1,52 +1 @@
-module.exports = loadAddon()
-
-// `require.addon()` reads the prebuild shipped inside this package. Since the
-// per-platform split (0.9.0) there is none, so the real binding almost always
-// comes from the `#host-addon` platform package. Runtimes have been seen to
-// answer `require.addon()` with this package's own JavaScript entry instead of
-// failing, which used to be returned verbatim: callers then got a module with
-// no `createInstance`, and the first symptom was an unrelated TypeError deep in
-// a model load. Anything that is not the native binding is treated as a miss.
-function loadAddon() {
-  let cause = null
-
-  try {
-    const addon = require.addon()
-    if (isNativeBinding(addon)) return addon
-    cause = new Error(
-      '@qvac/tts-ggml: require.addon() answered with a module that is not the native binding ' +
-        '(no createInstance function), so the prebuild in this package was treated as absent.'
-    )
-  } catch (err) {
-    cause = err
-  }
-
-  return loadPlatformPackageAddon(cause)
-}
-
-function loadPlatformPackageAddon(cause) {
-  let addon
-
-  try {
-    addon = require('#host-addon')
-  } catch (err) {
-    if (err.cause === undefined) err.cause = cause
-    throw err
-  }
-
-  if (!isNativeBinding(addon)) {
-    const err = new Error(
-      '@qvac/tts-ggml resolved #host-addon to a module that is not the native binding: ' +
-        'it has no createInstance function. Check that the platform package for this host ' +
-        'is installed and is not shadowed by another module of the same name.'
-    )
-    err.cause = cause
-    throw err
-  }
-
-  return addon
-}
-
-function isNativeBinding(addon) {
-  return addon !== null && typeof addon === 'object' && typeof addon.createInstance === 'function'
-}
+module.exports = require('#host-addon')

package/examples/moss-sfx-fit.js

@@ -8,11 +8,18 @@ if (!model || !prompt || seconds === undefined) {
   throw new Error('Usage: bare examples/moss-sfx-fit.js <model.gguf> <prompt> <seconds>')
 }
 
-const fit = TTSGgml.assessFit({
-  engineType: 'moss-sfx',
-  mossSoundEffectPath: model,
-  prompt,
-  seconds: Number(seconds),
-  useGPU: false
+async function main() {
+  const fit = await TTSGgml.assessFit({
+    engineType: 'moss-sfx',
+    mossSoundEffectPath: model,
+    prompt,
+    seconds: Number(seconds),
+    useGPU: false
+  })
+  console.log(fit.report || `${fit.status}: ${fit.reason}`)
+}
+
+main().catch((err) => {
+  console.error(err)
+  process.exit(1)
 })
-console.log(fit.report || `${fit.status}: ${fit.reason}`)

package/index.d.ts

@@ -1,6 +1,5 @@
 import { type QvacResponse } from "@qvac/infer-base";
 import * as errorModule from "./lib/error";
-import { resolveBackendsDir as resolveBackendsDirImpl } from "./lib/backends";
 import { assessFit as assessFitImpl, type Audio8FitRequest, type MossFitRequest, type MossSoundEffectFitRequest, type ChatterboxFitRequest, type CosyvoiceFitRequest, type ParlerFitRequest, type SupertonicFitRequest, type TtsFitEngine, type TtsFitRequest, type TtsFitResult, type TtsFitStatus } from "./lib/fit";
 import { type SentenceDelimiterPreset } from "./lib/textStreamAccumulator";
 declare const ENGINE_CHATTERBOX = "chatterbox";
@@ -607,7 +606,11 @@ interface TTSGgmlOptions extends ParlerDescriptionFields, Audio8VoiceFields, TTS
      * streaming (batch synthesis only).
      */
     denoiser?: LavaSRDenoiserOptions;
-    /** Directory the addon scans for dynamically loaded ggml backends. */
+    /**
+     * Directory the addon scans for dynamically loaded ggml backends, as given.
+     * Unset scans the addon's own: the one its platform package or the app
+     * installed next to the native module.
+     */
     backendsDir?: string;
     /** Directory where ggml-opencl persists its compiled program binary. */
     openclCacheDir?: string;
@@ -1211,7 +1214,6 @@ declare namespace TTSGgml {
     type MossFitRequest = NamespaceMossFit;
     type MossSoundEffectFitRequest = NamespaceMossSoundEffectFit;
     type CosyvoiceFitRequest = NamespaceCosyvoiceFit;
-    const resolveBackendsDir: typeof resolveBackendsDirImpl;
     const assessFit: typeof assessFitImpl;
 }
 export = TTSGgml;

package/index.js

@@ -46,7 +46,6 @@ const infer_base_1 = require("@qvac/infer-base");
 const tts_1 = require("./tts");
 const errorModule = __importStar(require("./lib/error"));
 const pocketConfig_1 = require("./lib/pocketConfig");
-const backends_1 = require("./lib/backends");
 const fit_1 = require("./lib/fit");
 const textChunker_1 = require("./lib/textChunker");
 const textStreamAccumulator_1 = require("./lib/textStreamAccumulator");
@@ -1293,7 +1292,7 @@ class TTSGgml {
         this._assignSynthesisOptions(options);
         this._enhancerGgufPath = resolveEnhancerGgufPath(normalizedFiles, options.enhancer);
         this._denoiserGgufPath = resolveDenoiserGgufPath(normalizedFiles, options.denoiser);
-        this._backendsDir = firstNonEmpty(options.backendsDir, this._config.backendsDir, (0, backends_1.resolveBackendsDir)());
+        this._backendsDir = firstNonEmpty(options.backendsDir, this._config.backendsDir);
         this._openclCacheDir = firstNonEmpty(options.openclCacheDir, this._config.openclCacheDir);
         this._vulkanCacheDir = firstNonEmpty(options.vulkanCacheDir, this._config.vulkanCacheDir);
         assertGpuIntentConsistent(this._config.useGPU, this._nGpuLayers);
@@ -3311,7 +3310,6 @@ class TTSGgml {
 (function (TTSGgml) {
     TTSGgml.QvacErrorAddonTTSGgml = errorModule.QvacErrorAddonTTSGgml;
     TTSGgml.ERR_CODES = errorModule.ERR_CODES;
-    TTSGgml.resolveBackendsDir = backends_1.resolveBackendsDir;
     TTSGgml.assessFit = fit_1.assessFit;
 })(TTSGgml || (TTSGgml = {}));
 module.exports.QvacErrorAddonTTSGgml =

package/lib/fit.d.ts

@@ -139,9 +139,9 @@ export interface TtsFitResult {
  *
  * `engineType` picks the fitter, the same key a load uses, and is required
  * here: a fit request carries none of the file keys a load is inferred from.
- * A model the engine cannot read comes back as `status: "error"`, as does a
+ * A model the engine cannot read resolves to `status: "error"`, as does a
  * voice with no fitter; a broken request, or a host with no native binding,
- * throws.
+ * rejects. Fits run one at a time within this addon.
  */
-export declare function assessFit(request: TtsFitRequest): TtsFitResult;
+export declare function assessFit(request: TtsFitRequest): Promise<TtsFitResult>;
 export {};

package/lib/fit.js

@@ -1,7 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.assessFit = assessFit;
-const backends_1 = require("./backends");
 /**
  * Projects one voice against the memory free right now, reading GGUF metadata
  * and never weight data. The registry's weightless copy of each file answers
@@ -13,17 +12,12 @@ const backends_1 = require("./backends");
  *
  * `engineType` picks the fitter, the same key a load uses, and is required
  * here: a fit request carries none of the file keys a load is inferred from.
- * A model the engine cannot read comes back as `status: "error"`, as does a
+ * A model the engine cannot read resolves to `status: "error"`, as does a
  * voice with no fitter; a broken request, or a host with no native binding,
- * throws.
+ * rejects. Fits run one at a time within this addon.
  */
-function assessFit(request) {
-    // eslint-disable-next-line @typescript-eslint/no-require-imports -- native binding is resolved lazily from package prebuilds.
+async function assessFit(request) {
+    // eslint-disable-next-line @typescript-eslint/no-require-imports -- the native binding loads on first use.
     const binding = require('../binding.js');
-    return binding.assessFit({
-        ...request,
-        backendsDir: typeof request.backendsDir === 'string' && request.backendsDir.length > 0
-            ? request.backendsDir
-            : (0, backends_1.resolveBackendsDir)()
-    });
+    return await binding.assessFit(request);
 }

package/package.json

@@ -1,8 +1,7 @@
 {
   "name": "@qvac/tts-ggml",
-  "version": "0.11.1",
+  "version": "0.13.0",
   "description": "Text to Speech (TTS) addon for qvac (ggml backend, wrapping the chatterbox + supertonic + parler + cosyvoice3 + audio8 + moss + moss-sfx + moss-speech + Pocket engines from tts-cpp)",
-  "addon": true,
   "qvac": {
     "ggml-speech": "2026-10-07#1"
   },
@@ -12,52 +11,32 @@
   "imports": {
     "#host-addon": {
       "linux": {
-        "x64": [
-          "@qvac/tts-ggml-linux-x64",
-          "./addon-unavailable.js"
-        ],
-        "arm64": [
-          "@qvac/tts-ggml-linux-arm64",
-          "./addon-unavailable.js"
-        ],
+        "x64": "@qvac/tts-ggml-linux-x64",
+        "arm64": "@qvac/tts-ggml-linux-arm64",
         "default": "./addon-unavailable.js"
       },
       "darwin": {
-        "arm64": [
-          "@qvac/tts-ggml-darwin-arm64",
-          "./addon-unavailable.js"
-        ],
-        "x64": [
-          "@qvac/tts-ggml-darwin-x64",
-          "./addon-unavailable.js"
-        ],
+        "arm64": "@qvac/tts-ggml-darwin-arm64",
+        "x64": "@qvac/tts-ggml-darwin-x64",
         "default": "./addon-unavailable.js"
       },
       "win32": {
-        "x64": [
-          "@qvac/tts-ggml-win32-x64",
-          "./addon-unavailable.js"
-        ],
+        "x64": "@qvac/tts-ggml-win32-x64",
         "default": "./addon-unavailable.js"
       },
       "android": {
-        "arm64": [
-          "@qvac/tts-ggml-android-arm64",
-          "./addon-unavailable.js"
-        ],
+        "arm64": "@qvac/tts-ggml-android-arm64",
         "default": "./addon-unavailable.js"
       },
-      "ios": [
-        "@qvac/tts-ggml-ios",
-        "./addon-unavailable.js"
-      ],
+      "ios": "@qvac/tts-ggml-ios",
       "default": "./addon-unavailable.js"
     }
   },
   "scripts": {
     "build": "npm run build:ts && npm run build:native",
-    "build:native": "bare-make generate -D BUILD_FUZZING=OFF && bare-make build && bare-make install",
-    "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON -D BUILD_FUZZING=OFF && bare-make build && bare-make install",
+    "build:native": "bare-make generate -D BUILD_FUZZING=OFF && bare-make build && bare-make install && npm run link:platform",
+    "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON -D BUILD_FUZZING=OFF && bare-make build && bare-make install && npm run link:platform",
+    "link:platform": "node ../../scripts/ci/slice-platform-packages.mjs --link-local --workdir .",
     "build:cuda": "npm run build:ts && npm run build:native:cuda",
     "build:ts": "tsc -p tsconfig.build.json",
     "build:pack": "npm run build:ts && mkdir -p dist && npm pack --pack-destination dist",
@@ -86,10 +65,10 @@
     "example:moss-sfx": "bare examples/moss-sound-effect.js \"Heavy rain falling on a tin roof with distant thunder.\" 5",
     "example:moss-speech": "bare examples/moss-speech.js test/reference-audio/jfk.wav",
     "lint": "npm run lint:js && npm run lint:ts && npm run test:types",
-    "lint:js": "prettier --check \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\" && lunte \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\"",
+    "lint:js": "prettier --check \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\" && lunte \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\"",
     "lint:ts": "eslint \"src/**/*.ts\" --max-warnings=0",
     "lint:ts:fix": "eslint \"src/**/*.ts\" --fix",
-    "format": "prettier --write \"test/**/*.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"binding.js\" \"addon-unavailable.js\"",
+    "format": "prettier --write \"test/**/*.js\" \"scripts/__tests__/package-contract.test.js\" \"binding.js\" \"addon-unavailable.js\"",
     "test:unit": "npm run build:ts && brittle-bare test/unit/**/*.test.js && npm run test:node",
     "test:node": "node --test scripts/__tests__/*.test.js",
     "test": "npm run test:types && npm run test:unit && npm run test:integration",
@@ -147,6 +126,11 @@
     "test/pocket-worklet",
     "test/reference-audio",
     "test/utils/runTTS.js",
+    "test/utils/mossBackend.js",
+    "test/utils/mossGpuValidation.js",
+    "test/utils/gpuOnly.js",
+    "test/utils/testThreads.js",
+    "test/utils/perf-helper.js",
     "test/utils/runChatterboxTTS.js",
     "test/utils/runSupertonicTTS.js",
     "test/utils/runParlerTTS.js",
@@ -221,7 +205,7 @@
   },
   "dependencies": {
     "@qvac/error": "^0.1.0",
-    "@qvac/infer-base": "^0.6.2",
+    "@qvac/infer-base": "^0.8.0",
     "@qvac/langdetect-text": "^0.1.2",
     "@qvac/logging": "^0.1.0",
     "bare-fs": "^4.5.6",
@@ -232,12 +216,24 @@
     "bare-stream": "^2.7.0",
     "bare-url": "^2.4.3"
   },
-  "types": "index.d.ts",
   "optionalDependencies": {
-    "@qvac/tts-ggml-linux-x64": "0.11.1",
-    "@qvac/tts-ggml-linux-arm64": "0.11.1",
-    "@qvac/tts-ggml-darwin-arm64": "0.11.1",
-    "@qvac/tts-ggml-darwin-x64": "0.11.1",
-    "@qvac/tts-ggml-win32-x64": "0.11.1"
+    "@qvac/tts-ggml-linux-x64": "0.13.0",
+    "@qvac/tts-ggml-linux-arm64": "0.13.0",
+    "@qvac/tts-ggml-darwin-arm64": "0.13.0",
+    "@qvac/tts-ggml-darwin-x64": "0.13.0",
+    "@qvac/tts-ggml-win32-x64": "0.13.0"
+  },
+  "types": "index.d.ts",
+  "peerDependencies": {
+    "@qvac/tts-ggml-android-arm64": "0.13.0",
+    "@qvac/tts-ggml-ios": "0.13.0"
+  },
+  "peerDependenciesMeta": {
+    "@qvac/tts-ggml-android-arm64": {
+      "optional": true
+    },
+    "@qvac/tts-ggml-ios": {
+      "optional": true
+    }
   }
 }