May I ask how I can use tflite/sensor flow more effectively on Snapdragon processors?

General Discussion
,
#1

Recently, I am trying to deploy my model on an Android phone. I learned about Snapdragon’s powerful AI computing capabilities through the internet, but I am unable to find the correct way to use this powerful feature. May I ask if someone has done something related to it? Can you give me some guidance? thanks

#3

Hi @jun_yin1, You can use Delegates, which enable hardware acceleration of TensorFlow Lite models by leveraging on-device accelerators.

  • You can use NNAPI delegates to accelerate models on Android devices with GPU, DSP and / or NPU available. It is available in Android 8.1 (API 27+) or higher.
  • The Hexagon delegate can be used to accelerate models on Android devices with Qualcomm Hexagon DSP.

For more details please refer to this document. Thank You.

#5

Thank you for your reply. Also, if I wish to use TensorFlow/Pytorch to directly use Snapdragon’s acceleration function, can I?

#7

Hi @jun_yin1, No, these delegates can be used for tflite only. Thank You.