Android ML's early access program gives developers the opportunity to partner with the Android ML team and have access to upcoming APIs that are still under development. This collaboration allows us to optimize the implementations before we make them widely available in the official release.
We make decisions on the release of an API based on the feedback we receive during the early access program.
The early access program also lets developers help shape the design of new APIs by giving us feedback on the onboarding experience, thoughts on features, and suggestions on how it can be used. We make decisions on the release of an API partially based on your feedback.
Early access APIs
We are accepting applications for early access to these APIs:
Acceleration Service for Android: A new API that allows selecting the optimal hardware accelerator for running TensorFlow Lite inference at runtime. Acceleration Service works by running local benchmarks on user devices to determine the optimal accelerator and minimizes variations caused by hardware differences.
Updatable Hardware Drivers: Allows benefitting from the latest optimized drivers and TensorFlow Lite delegates for hardware acceleration.
Fill out this form to join our early access program and request access to these APIs.
If your request is accepted, a member of the Android ML team will contact you and give a detailed document outlining the EAP terms and usage instructions. Depending on the particular API, you might have to complete additional forms to sign in accordance with the program.
The Android ML team will stay in touch to establish the steps for providing feedback once you start using the API.
TensorFlow Lite in Google Play services
The play services API for TensorFlow Lite, which used to be only available via our early access program, is now publicly available in beta.
Using this API rather than regular TensorFlow Lite allows you to reduce your apk size and benefit from regular updates to TensorFlow Lite runtime, which may translate into performance improvements.
You can access the developer documentation here.