On October 28, Google announced CompiledModel API and ML Drift GPU inference for the shared LiteRT runtime in Play services. It avoids bundling the runtime with each app and describes CPU fallback and continued Interpreter CPU support. The post also introduces migration through the litert-runtime skill in Android Studio Agent Mode. Developers should review changes and check outputs and compatibility on real devices. Performance and size improvements are vendor measurements; NPU support remains in development.
0 comments
Sign in · Sign in to leave a comment.