QVAC-18481 test[skiplog]: pre-load multi-file model companion sets on bootstrap by Victor-Rodzko · Pull Request #1899 · tetherto/qvac

Victor-Rodzko · 2026-05-05T11:52:58Z

Note: be concise and prefer bullet points.

🎯 What problem does this PR solve?

tts-chatterbox-short-text flakes on Android smoke (first test cold-loads ~5 ONNX files + ONNX Runtime init, can exceed the 600s test watchdog)
bootstrap() only pre-downloaded the main constant for multi-file models (chatterbox, supertonic, parakeet, vision); companion files were lazily fetched during the first test
Mobile chatterbox needs referenceAudioSrc resolved from a bundled RN asset URI — can't be expressed as a static config object
A few iOS transcribe() call sites slipped past the existing ^transcription- skip regex (config-reload-then-transcribe, error-transcription-failed)

📝 How does it solve it?

Switch multi-file model resources to preLoadUnload: true on mobile + desktop. loadModel() triggers downloadCompanionSetFromRegistry() which fetches the entire set; the subsequent unloadModel() keeps the file cache + warm native paths
Drop redundant skipPreDownload: true so models download in parallel during the download phase, then pre-load runs sequentially after
Add async config resolver support in ResourceManager (cached per-dep). Mobile chatterbox uses it to resolve referenceAudioSrc from expo-asset at bootstrap time
Remove obsolete patchChatterboxReferenceAudio() workaround in MobileTtsExecutor — config now carries everything before the first test runs
Extract shared resolveBundledAssetUri helper (mobile/asset-uri.ts) — model-asset-executor.ts:resolveAsset() and the chatterbox audio lookup now delegate to it (was a duplicated 12-line block in two places)
Extend iOS transcribe() skip list to cover config-reload-then-transcribe and error-transcription-failed

🧪 How was it tested?

Local smoke on macOS: 89/89 passed, 4:20 min, all 8 preLoadUnload models pre-cached cleanly
Local smoke on iPhone 16e (iOS): 76/89 passed + 13 skipped, 0 failed, 4:58 min, peak memory 2.45 GB; tts-chatterbox-short-text green at 15.8s
Verified bootstrap log shows full companion-set fetches: chatterbox pulls all 5 ONNX files at pre-load (vs only tokenizer.json previously)

Switch chatterbox/supertonic/parakeet/vision/diffusion to preLoadUnload so loadModel() at bootstrap fetches every companion file (encoder, decoder, vocab, projection, etc.) — otherwise they were lazily fetched inside the first test, which caused the tts-chatterbox-short-text Android flake (5 ONNX files + ONNX Runtime cold init blew through the 600s test watchdog). Add async config resolver to ResourceManager so chatterbox can resolve its referenceAudioSrc from the bundled RN asset registry at bootstrap time; cached per-dep. Remove the now-obsolete patchChatterboxReferenceAudio workaround from MobileTtsExecutor. Extend the iOS transcribe() skip list to catch the call sites that slipped past the ^transcription- regex (config-reload-then-transcribe, error-transcription-failed). Co-authored-by: Cursor <cursoragent@cursor.com>

Pull asset resolution + file:// stripping out of the duplicated copies in mobile/consumer.ts and mobile/executors/model-asset-executor.ts into a single mobile/asset-uri.ts helper. Both sites now delegate, so future changes to expo-asset handling live in one place. Tighter idioms in the helper itself: regex strip instead of substring(7), ?? instead of ||, no mutable let. Co-authored-by: Cursor <cursoragent@cursor.com>

Add `clean:sdk-snapshot` to wipe the cached @qvac/sdk copies left over in tests-qvac/node_modules and the iOS/Android consumer build dirs by previous `npm install --install-links` runs. Wire it into `install:build:full` so a full rebuild always pulls a fresh SDK snapshot after `prepare:sdk` rebuilds the SDK itself. Co-authored-by: Cursor <cursoragent@cursor.com>

github-actions · 2026-05-05T12:15:44Z