@qvac/audiogen-ggml 0.7.0 → 0.9.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -5,6 +5,68 @@ All notable changes to this project will be documented in this file.
5
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
6
6
  and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
7
 
8
+ ## [0.9.0] - 2026-10-23
9
+
10
+ ### Added
11
+
12
+ - Bounded FuzzTest coverage for the JS-adapter number parsers
13
+ (`parseInteger`, `parseFloat`, `checkedInteger`, `checkedFloat`)
14
+ ([#4606](https://github.com/tetherto/qvac/pull/4606)). Linux C++ CI runs the
15
+ suite after the native build's caches are saved; this addon has no
16
+ addon-level C++ unit tests yet. The parsers compile without audiogen-cpp, so
17
+ ASan and LeakSanitizer stay at full strength. No public addon API changes.
18
+
19
+ ### Changed
20
+
21
+ - Align the shared native `speech-cpp` source pin to `2026-10-20#1` across
22
+ the speech addons so they use the same source revision
23
+ ([#5062](https://github.com/tetherto/qvac/pull/5062),
24
+ [#5194](https://github.com/tetherto/qvac/pull/5194),
25
+ [#5245](https://github.com/tetherto/qvac/pull/5245)).
26
+ - Move the shared native `speech-cpp` source pin to `2026-10-23#2` across
27
+ the speech addons so they use the same source revision
28
+ ([#5326](https://github.com/tetherto/qvac/pull/5326)).
29
+
30
+ ### Fixed
31
+
32
+ - Adreno GPUs (Android, OpenCL): q8_0 matrix products whose sums exceed
33
+ the fp16 range no longer come back as infinity, and attention over 32 or
34
+ more tokens with unpadded lengths is computed correctly. The shared native
35
+ `ggml-speech` moves to `2026-10-23` across the speech addons
36
+ ([#5326](https://github.com/tetherto/qvac/pull/5326)).
37
+
38
+ ## [0.8.0] - 2026-10-14
39
+
40
+ ### Breaking
41
+
42
+ - `assessFit` returns a Promise; the synchronous binding is removed
43
+ ([#5010](https://github.com/tetherto/qvac/pull/5010)). The fit runs on a worker thread so the JavaScript loop keeps
44
+ running, and concurrent calls run one at a time within the addon. A model
45
+ the fitter cannot read still resolves with `status: 'error'`; a malformed
46
+ request rejects instead of throwing. Change
47
+ `const fit = assessFit(request)` to `const fit = await assessFit(request)`.
48
+
49
+ ### Changed
50
+
51
+ - `@qvac/infer-base` dependency raised `^0.6.2` -> `^0.8.0`
52
+ ([#5065](https://github.com/tetherto/qvac/pull/5065)). `^0.6.2` never took 0.7.0 or 0.8.0; 0.8.0 adds an
53
+ optional `retainOutput: false` on `QvacResponse` and
54
+ `createJobHandler().start()` for consuming output events through
55
+ `onUpdate()` without keeping every chunk in memory ([#5028](https://github.com/tetherto/qvac/pull/5028)).
56
+ Retention stays enabled by default and this package does not pass the option, so there is no API or
57
+ runtime behaviour change for this package.
58
+ - `speech-cpp` is pinned to `2026-10-12` port-version 1, the registry
59
+ revision that ships the merged MOSS CUDA/Vulkan backend source, the same
60
+ pin as the other speech addons ([#5001](https://github.com/tetherto/qvac/pull/5001)).
61
+
62
+ ### Fixed
63
+
64
+ - The linux-x64 prebuild ships the CUDA backend again, next to Vulkan and
65
+ the CPU variants; releases 0.5.0 to 0.7.0 left it out. Where the NVIDIA
66
+ driver and the CUDA 13 runtime libraries (cudart, cuBLAS) resolve, GPU
67
+ requests run on CUDA instead of Vulkan; other hosts keep Vulkan or CPU
68
+ ([#4985](https://github.com/tetherto/qvac/pull/4985)).
69
+
8
70
  ## [0.7.0] - 2026-10-14
9
71
 
10
72
  ### Breaking
package/README.md CHANGED
@@ -709,10 +709,12 @@ answers the same as the file itself and the projection can run before anything
709
709
  is downloaded. It is a module export, not an instance method — nothing is
710
710
  loaded to call it.
711
711
 
712
+ `assessFit` returns a Promise. Fits run one at a time within the addon.
713
+
712
714
  ```js
713
715
  const { assessFit } = require('@qvac/audiogen-ggml')
714
716
 
715
- const fit = assessFit({
717
+ const fit = await assessFit({
716
718
  modelsDir: '/models/ace-step',
717
719
  durationSeconds: 30,
718
720
  textTokens: 256,
@@ -760,7 +762,7 @@ peak of repeated generations, and the `stages` rows are `lm`, `depth`, `cond`,
760
762
  `dit` and `vocoder`.
761
763
 
762
764
  ```js
763
- const fit = assessFit({
765
+ const fit = await assessFit({
764
766
  engine: 'minimax',
765
767
  modelsDir: '/models/minimax-music3',
766
768
  device: 'gpu',
@@ -780,7 +782,7 @@ On Android and iOS, where MiniMax-Music3 is unavailable, `engine: 'minimax'`
780
782
  comes back as `status: "error"` with `reason: "unsupported-engine"`.
781
783
 
782
784
  A model the engine cannot read is `status: "error"`; a broken request, or a
783
- host with no native binding, throws.
785
+ host with no native binding, rejects.
784
786
 
785
787
  ## Lifecycle
786
788
 
@@ -833,6 +835,28 @@ dispatches to the corresponding `audiogen-cpp` engine.
833
835
  them keeps one copy of each: use releases that record the same
834
836
  `ggml-speech`.
835
837
 
838
+ ## Fuzzing
839
+
840
+ The JS-adapter number parsers have a [Google FuzzTest][fuzztest] target,
841
+ `audiogen-config-parse-fuzz`. Fuzzing is Linux-only: it needs clang with
842
+ libFuzzer and AddressSanitizer, CMake 3.25 or newer, [vcpkg](https://vcpkg.io/)
843
+ (set `VCPKG_ROOT`) and `bare-make`. The target does not link audiogen-cpp, so
844
+ it runs with full ASan and LeakSanitizer.
845
+
846
+ ```bash
847
+ npm install
848
+ npm run fuzz # bounded run of every FUZZ_TEST, as Linux CI does
849
+ npm run fuzz:continuous # coverage-guided, one FUZZ_TEST at a time
850
+ npm run fuzz:continuous -- AudiogenConfigParseFuzz.ParseFloatNeverCrashes --fuzz_for=30m
851
+ ```
852
+
853
+ Flags other than `--continuous` and `--build-dir` go to the fuzz binary. See
854
+ [`docs/architecture/ADDON-FUZZING.md`](../../docs/architecture/ADDON-FUZZING.md)
855
+ for the fuzzing design, the vcpkg-supplied FuzzTest stack, and how to add a
856
+ target.
857
+
858
+ [fuzztest]: https://github.com/google/fuzztest
859
+
836
860
  ## Benchmarking
837
861
 
838
862
  The package ships the Real-Time Factor benchmark it is measured with, so the
package/lib/fit.d.ts CHANGED
@@ -113,4 +113,4 @@ export interface AudiogenFitResult {
113
113
  report: string
114
114
  }
115
115
 
116
- export function assessFit(request: AudiogenFitRequest): AudiogenFitResult
116
+ export function assessFit(request: AudiogenFitRequest): Promise<AudiogenFitResult>
package/lib/fit.js CHANGED
@@ -11,12 +11,13 @@
11
11
  * `status: "error"` with `reason: "unsupported-engine"`.
12
12
  *
13
13
  * A model set the engine cannot read comes back as `status: "error"` with its
14
- * own reason; a broken request, or a host with no native binding, throws.
14
+ * own reason; a broken request, or a host with no native binding, rejects.
15
+ * Fits run one at a time within the addon.
15
16
  */
16
- function assessFit(request) {
17
+ async function assessFit(request) {
17
18
  const binding = require('../binding.js')
18
19
 
19
- return binding.assessFit(request)
20
+ return await binding.assessFit(request)
20
21
  }
21
22
 
22
23
  module.exports = { assessFit }
package/package.json CHANGED
@@ -1,9 +1,9 @@
1
1
  {
2
2
  "name": "@qvac/audiogen-ggml",
3
- "version": "0.7.0",
3
+ "version": "0.9.0",
4
4
  "description": "Audio generation (music) addon for qvac",
5
5
  "qvac": {
6
- "ggml-speech": "2026-10-07#1"
6
+ "ggml-speech": "2026-10-23"
7
7
  },
8
8
  "engines": {
9
9
  "bare": ">=1.19.0"
@@ -34,8 +34,8 @@
34
34
  },
35
35
  "scripts": {
36
36
  "build": "npm run build:ts && npm run build:native",
37
- "build:native": "bare-make generate && bare-make build && bare-make install && npm run link:platform",
38
- "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON && bare-make build && bare-make install && npm run link:platform",
37
+ "build:native": "bare-make generate -D BUILD_FUZZING=OFF && bare-make build && bare-make install && npm run link:platform",
38
+ "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON -D BUILD_FUZZING=OFF && bare-make build && bare-make install && npm run link:platform",
39
39
  "link:platform": "node ../../scripts/ci/slice-platform-packages.mjs --link-local --workdir .",
40
40
  "build:cuda": "npm run build:ts && npm run build:native:cuda",
41
41
  "build:ts": "tsc -p tsconfig.build.json",
@@ -59,6 +59,11 @@
59
59
  "test:integration": "npm run build:ts && brittle-bare test/integration/**/*.test.js",
60
60
  "test:benchmark:rtf": "npm run build:ts && brittle-bare test/benchmark/rtf-benchmark.test.js",
61
61
  "test:benchmark:rtf:matrix": "node scripts/run-rtf-benchmark-matrix.js",
62
+ "fuzz:build": "bare-make generate -D BUILD_FUZZING=ON -D BUILD_TESTING=OFF -D FUZZTEST_FUZZING_MODE=OFF && bare-make build --target audiogen-config-parse-fuzz",
63
+ "fuzz:build:continuous": "bare-make generate -D BUILD_FUZZING=ON -D BUILD_TESTING=OFF -D FUZZTEST_FUZZING_MODE=ON && bare-make build --target audiogen-config-parse-fuzz",
64
+ "fuzz:run": "node scripts/run-cpp-fuzz.js",
65
+ "fuzz": "npm run fuzz:build && npm run fuzz:run",
66
+ "fuzz:continuous": "npm run fuzz:build:continuous && node scripts/run-cpp-fuzz.js --continuous",
62
67
  "test": "npm run test:types && npm run test:unit && npm run test:integration",
63
68
  "lint": "npm run lint:js && npm run lint:ts && npm run test:types",
64
69
  "lint:js": "prettier --check \"addon-unavailable.js\" \"binding.js\" \"examples/*.js\" \"lib/*.js\" \"scripts/**/*.{js,mjs}\" \"test/**/*.js\" && lunte \"addon-unavailable.js\" \"binding.js\" \"examples/*.js\" \"lib/*.js\" \"scripts/**/*.{js,mjs}\" \"test/**/*.js\"",
@@ -126,7 +131,7 @@
126
131
  ],
127
132
  "devDependencies": {
128
133
  "@qvac/registry-client": "^0.6.1",
129
- "bare-pack": "^2.2.3",
134
+ "bare-pack": "^3.0.0",
130
135
  "brittle": "^3.17.0",
131
136
  "cmake-bare": "^1.7.5",
132
137
  "cmake-vcpkg": "^1.1.0",
@@ -142,7 +147,7 @@
142
147
  },
143
148
  "dependencies": {
144
149
  "@qvac/error": "^0.1.1",
145
- "@qvac/infer-base": "^0.6.2",
150
+ "@qvac/infer-base": "^0.8.0",
146
151
  "@qvac/logging": "^0.1.0",
147
152
  "bare-ffmpeg": "^1.4.0",
148
153
  "bare-fs": "^4.5.6",
@@ -151,19 +156,19 @@
151
156
  "bare-process": "^4.2.2"
152
157
  },
153
158
  "optionalDependencies": {
154
- "@qvac/audiogen-ggml-linux-x64": "0.7.0",
155
- "@qvac/audiogen-ggml-linux-arm64": "0.7.0",
156
- "@qvac/audiogen-ggml-darwin-arm64": "0.7.0",
157
- "@qvac/audiogen-ggml-darwin-x64": "0.7.0",
158
- "@qvac/audiogen-ggml-win32-x64": "0.7.0"
159
+ "@qvac/audiogen-ggml-linux-x64": "0.9.0",
160
+ "@qvac/audiogen-ggml-linux-arm64": "0.9.0",
161
+ "@qvac/audiogen-ggml-darwin-arm64": "0.9.0",
162
+ "@qvac/audiogen-ggml-darwin-x64": "0.9.0",
163
+ "@qvac/audiogen-ggml-win32-x64": "0.9.0"
159
164
  },
160
165
  "overrides": {
161
166
  "bare-module-lexer": "1.6.3"
162
167
  },
163
168
  "types": "index.d.ts",
164
169
  "peerDependencies": {
165
- "@qvac/audiogen-ggml-android-arm64": "0.7.0",
166
- "@qvac/audiogen-ggml-ios": "0.7.0"
170
+ "@qvac/audiogen-ggml-android-arm64": "0.9.0",
171
+ "@qvac/audiogen-ggml-ios": "0.9.0"
167
172
  },
168
173
  "peerDependenciesMeta": {
169
174
  "@qvac/audiogen-ggml-android-arm64": {