build / build (push) Successful in 16m0s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
731 B
731 B
Stubs of the native ML packages (tth fork)
pubspec.yaml substitutes them through dependency_overrides for
llamadart, flutter_onnxruntime and flutter_langdetect. They carry only
the API the app calls, and every call throws UnsupportedError. The native
llama.cpp / LiteRT-LM / ONNX Runtime libraries (~200 MB) therefore never get
into the APK, and the ML code in lib/ stays as it is upstream.
Consequence: message translation and the neural image codec do not work.
Both are off by default upstream, the app catches and logs the error, and the
fork hides their settings (kOnDeviceMl in lib/tth_fork.dart).
When upstream starts calling another symbol, the build fails on the missing identifier — add it here.