I bought a Coral Edge TPU for my Raspberry Pi to use TensorFlow Lite.
On provider's homepage, They said it's only for inferencing and just limited transfer learning.
There example use it their own framework. Is Tensorflow-core library support this device?
Colar Edge TPU designed for inferencing. There is no FP32 or FP16 operation units. And provider's library also don't have it. So it's impossible what i want to do.
and Tensorflow-core library also do not support this device.
The EdgeTPU can only be used for tflite models that are compiled using the edgetpu_compiler. The job of the compiler is to map the model to the tpu, otherwise all operations will be executes on the CPU by default. The EdgeTPU, then, will not work with tensorflow core models because it hasn't been compiled. In order for you model(s) to pass the compiler, it will have to meets all requirements lists here.