1

[Solve]

  1. I found a post on edgetpu officail github, flex delegate is still unsupported.

  2. Another post also metioned that 5D transpose operation is still unsupported; Currently there are no plans in the future to support transpose op for tensor with dimensions > 3.

  • Hope these informations will be useful for those who see this post in the future.

Compilation error with edgetpu compiler

I'm trying to compile my fully quantized tflite model to edgetpu version and got the error messages as shown below.

Note that

  1. the model was DeiT, a vision trasfomer model released by Facebook, and I converted it via pytorch -> onnx -> tensorflow.

  2. I use edgetpu compiler under Google Colab environment.

Edge TPU Compiler version 16.0.384591198
Started a compilation timeout timer of 180 seconds.
ERROR: Regular TensorFlow ops are not supported by this interpreter. Make sure you apply/link the Flex delegate before inference.
ERROR: Node number 68 (FlexErf) failed to prepare.

Compilation failed: Model failed in Tflite interpreter. Please ensure model can be loaded/run in Tflite interpreter.
Compilation child process completed within timeout period.
Compilation failed!

Fail to run model through tflite-runtime

  1. I correctly run inference through tf.lite.Interpreter (version: 2.8.0) on my RaspberryPi 4 (aarch64).

  2. But I can't run inference successfuly through tflite-runtime (version: 2.9.1), the package of a subset of tensorflow to run inference only.

RuntimeError: Select TensorFlow op(s), 
included in the given model, is(are) not supported by this interpreter. 
Make sure you apply/link the Flex delegate before inference. 
For the Android, it can be resolved by adding "org.tensorflow:tensorflow-lite-select-tf-ops" dependency. 
See instructions: https://www.tensorflow.org/lite/guide/ops_selectNode number 68 (FlexErf) failed to prepare.
Select TensorFlow op(s), included in the given model, is(are) not supported by this interpreter. Make sure you apply/link the Flex delegate before inference.
For the Android, it can be resolved by adding "org.tensorflow:tensorflow-lite-select-tf-ops" dependency. 
See instructions: https://www.tensorflow.org/lite/guide/ops_selectNode number 68 (FlexErf) failed to prepare.

There are similar errors point to Flex delegate.

Here's my question,

  • And how to resovle Coral TPU edgetpu compiler error?

Since I'm a noob in deep learning, any suggestion will be helpful for me!

Thank you for your patience!

Heyward
  • 11
  • 2

0 Answers0