Smaller apps and faster GPU inference with LiteRT in Google Play services

LiteRT in Google Play services, Google’s on-device runtime for custom Android machine-learning models, has introduced GPU inference through its CompiledModel API, delivering approximately 1.4 times faster performance than the legacy TensorFlow Lite GPU delegate. By using the Play services-hosted runtime rather than bundling it within each app, developers can reduce app sizes by about 8 MB, receive automatic performance, security, and bug-fix updates, and support Android devices from Android 7 onward as well as recent Wear OS and Android Automotive versions. The CompiledModel API retains compatibility with .tflite models while simplifying CPU and GPU acceleration, using OpenCL or OpenGL with CPU fallback when needed, and offering cross-platform portability. Google recommends that developers migrate from deprecated legacy GPU and custom delegate APIs, while noting that existing Interpreter API code remains supported for CPU use; Android Studio’s new LiteRT skill can assist with migration. Future work will expand GPU support and add NPU support through Play services.

No tracked trend matches for this post yet.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.