Skip to content

Releases: openvinotoolkit/tflite_openvino_delegate

Intel OpenVINO NPU Tensorflow Lite Stable Delegate

11 Jun 18:22
@anikulk anikulk
v0.1
9903ffe
Compare
Choose a tag to compare
Intel OpenVINO NPU Tensorflow Lite Stable Delegate Pre-release
Pre-release

ChromeOS web applications and OS features currently access AI accelerators like NPU (Neural Processing Unit) via the Intel OpenVINO Delegate for TFLite.
We also have CPU delegates which can be tested via the Tensorflow Lite Delegate for OpenVINO.

We implement a stable opaque tflite delegate (more info about stable delegates can be found here ) for OpenVINO. We leverage the SimpleOpaqueDelegateKernelInterface and SimpleOpaqueDelegateInterface provided by Tflite.

At a high level we have the following modules

  • openvino_delegate_core : interface into OpenVINO (model, inferrequest )
  • openvino_graph_builder : Builds the ov::Model from tflite graph
  • operations : logic to translate tflite ops to ov ops

Maintainers/Code owners:
Ritul Jasuja ritul.jasuja@intel.com
Anisha Dattatraya Kulkarni anisha.dattatraya.kulkarni@intel.com
Vijeetkumar Benni vijeetkumar.benni@intel.com

Release notes for Release 0.1 :

Current release is alpha release for the TFLite Delegate. It is utilized by chromium OS and code can be found here

  • We enable a short list of models needed for Video Conferencing use cases in ChromeOS and the underlying ops needed.
  • With TFlite Delegate NNHAL support for NPU and OpenVINO is deprecated.

Contributions:

Loading