Closed enricoschroeder closed 1 year ago
@enricoschroeder our recommendation is to use tflite-runtime 2.5.0.post1. If you would like use with tf 2.10, 2.12 or any other version, please build libedgetpu, pycoral and tflite_runtime pacakges for your platform, using the instructions at: https://coral.ai/docs/notes/build-coral/. Thanks!!
@hjonnala Thanks for the answer! Before I go down the rabbit hole to build everthing with a current tensorflow version: Am I right to expect that using the regular tensorflow.lite.Interpreter with a matching libedgetpu should work?
Also, is there any particular reason why you do not provide more up-to-date versions of everything?
@hjonnala Thanks for the answer! Before I go down the rabbit hole to build everthing with a current tensorflow version: Am I right to expect that using the regular tensorflow.lite.Interpreter with a matching libedgetpu should work?
Yes..
Description
I am trying to load an *edgetpu.tflite model (for example the supplied mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite or any other model I tried) with the regular tensorflow.lite.Interpreter and the edgetpu delegate. This silently crashes (I suspect a segmentation fault in the background).
I tried with tensorflow 2.10.0 and 2.12.0. I can load the same model on the same machine with tflite-runtime 2.5.0 without any problems, however I need regular Tensorflow for dataset operations in my code and both (tflite-runtime and regular tensorflow) cannot be imported at the same time. Since the tflite Interpreter is included in the regular tensorflow package (and I have seen recommendations of using it instead of the dedicated tflite-runtime package), I would assume that this should work, shouldn't it?
Click to expand!
### Issue Type Bug ### Operating System Windows 10 ### Coral Device USB Accelerator ### Other Devices _No response_ ### Programming Language Python 3.9 ### Relevant Log Output _No response_