LiteRT delegate for Pixel Phones (Soc Tensor G2/G3/G4) TPUs

[caofx0418]

Before Android 15, LiteRT nnapi delegate could accelerate NN models.

But after the release of Android 15, NNAPI will be deprecated.

Compare to Qualcomm SOC, it offers QNN TFlite Delegate.

Will the Pixel Phones offer TPU delegate ?

[gaikwadrahul8]

Hi, @caofx0418
I apologize for the delay in my response, As far I know currently there is not a publicly available separate TPU delegate for TFLite specifically for Pixel's Tensor TPUs in the same way Qualcomm provides QNN. If I'm not wrong at the moment TFLite GPU delegate is the most likely path to hardware acceleration on Pixel's Tensor TPUs after NNAPI deprecation . I would suggest you to please refer this official documentation about NNAPI Migration Guide

Thank you for your cooperation and patience.

[caofx0418]

Hi, @caofx0418 I apologize for the delay in my response, As far I know currently there is not a publicly available separate TPU delegate for TFLite specifically for Pixel's Tensor TPUs in the same way Qualcomm provides QNN. If I'm not wrong at the moment TFLite GPU delegate is the most likely path to hardware acceleration on Pixel's Tensor TPUs after NNAPI deprecation . I would suggest you to please refer this official documentation about NNAPI Migration Guide

Thank you for your cooperation and patience.

Thank you for your response, but in "https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/experimental/litert/vendors/google_tensor", it seems to supply a TPU delegate as Qualcomm QNN and Mtk NeuroPilot

[caofx0418]

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/experimental/litert/vendors

[gaikwadrahul8]

Hi, @caofx0418
I apologize for the delay in response, Please refer this official documentation and as per that official documentation

We look forward to supporting delegates from the following vendors in the coming months:

• Google Pixel
• MediaTek
• Samsung System LSI

Stay tuned for updates and further instructions on using these delegates to harness the power of NPUs in your TFLite models.
If I'm not wrong it's in progress according to official documentation. The timeline is described as in the coming months. Meanwhile you can continue using NNAPI in Android 14 and below

Thank you for your cooperation and patience.

[gaikwadrahul8]

Hi, @caofx0418
If your question has been answered and the issue is resolved, please feel free to close this issue and for future updates please refer to the official documentation or the following link:https://developers.googleblog.com/en/search/?author=Google+AI+Edge+team

Thank you for your cooperation and understanding.

[caofx0418]

thank you for you response!