experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
rei/fork-r2.3
STT-tensorflow/tensorflow/lite/experimental/acceleration/README.md
A. Unique TensorFlower 5254b55e35 Add C++ and Java APIs for GPU delegate whitelisting. 
The GPU delegate is not supported on all Android devices due to differences in
OpenGL ES versions, driver capabilities and stability. This change provides a
whitelist that can be used to detect whether the delegate is supported on the
current device.

PiperOrigin-RevId: 314495681
Change-Id: I86c444188ebf998d6cfb1ea27428ce0900e926db
2020-06-03 02:45:50 -07:00

510 B
Raw Permalink Blame History

Accelerator whitelisting

Experimental library and tools for determining whether an accelerator engine works well on a given device, and for a given model.

Platform-agnostic, Android-first

Android-focused, since the much smaller set of configurations on iOS means there is much less need for whitelisting on iOS.

Not just for TfLite

This code lives in the TfLite codebase, since TfLite is the first open-source customer. It is however meant to support other users (direct use of NNAPI, mediapipe).