On-device-training using tflite_runtime wheel built based tf2.7

TensorFlow General Discussion
, , ,
1

Hello,

I’m trying to explore the possibilities of launching on device training on a raspberry 3B+ using the tflite_runtime python wheel that I have built manually based on the release 2.7 which says it supports on-device-training. I’m following currently the examples provided by these links:
On-Device Training with TensorFlow Lite
On-device training in TensorFlow Lite — The TensorFlow Blog
The Python examples starts by importing tensorflow which is not available on my board, I was wondering if there is a way to implement the same example but with the tflite_runtime.
Any help or examples would be appreciated.

Thank you in advance,

3

The on-device part start from “Convert model to TensorFlow Lite format” section in:

5

Thank your @Bhack for your answer. The only example given in the link you sent is written in Java, do you know if there any similar examples in C++ ?

6

You could try to ask if it is planned at:

https://github.com/tensorflow/community/pull/390