Internal compiler error. Aborting! with edgetpu_compiler

[akanksharma]

We are getting the below mentioned error with one of our quant models created with TF2.4.

# edgetpu_compiler ufs_128_quant.tflite
Edge TPU Compiler version 14.1.317412892

Internal compiler error. Aborting! 

We have tried to reduce the model size to around 5 MB only but the issue still exists.
We are however able to compile all other models except this one.
Unfortunately there are no logs available.

Any help to solve this issues is appreciated.

There is one more similar issue created about this but not sure if the actual cause is same.
#189

Thanks in advance.

[hjonnala]

please try using new compiler version. If it won't work, please share the ufs_128_quant.tflite model.

[akanksharma]

I tried to upgrade but still I am getting the same error

ufs_128_quant.tflite.zip

# edgetpu_compiler ufs_128_quant.tflite
Edge TPU Compiler version 15.0.340273435

Internal compiler error. Aborting!

Attached the quant model here.

[hjonnala]

The compiler is failing as there are too many of transpose ops which are not supported.

You can use https://netron.app/ to visualize the models.

please check all supported operations here: https://coral.ai/docs/edgetpu/models-intro/#supported-operations

[letdivedeep]

Hi @hjonnala

I am trying to convert PyTorch model to tflite using this pipeline :
Pytorch -> onnx -> saved_model -> tflite

As we know Pytorch format is (NCHW) and the PB from is (NHWC), so when we convert using onnx2tf it adds lots of transpose to handle this

do we have an alternative way to avoid it and make it run on edge tpu

[hjonnala]

@letdivedeep switching the model architecture might resolve the issue.
Transpose operation could be supported be in future releases.

[hjonnala]

@ akanksharma are you able to modify the model that can compiled with edgetpu compiler?

[letdivedeep]

@hjonnala

we were able to resolve the issue by having a workaround with openvino2tensorflow :

Pytorch -> onnx -> openViNO -> edge_tpu

https://github.com/PINTO0309/openvino2tensorflow

blog: https://qiita.com/PINTO/items/ed06e03eb5c007c2e102

[hjonnala]

@letdivedeep Thanks for sharing the workaround links.

[taloot]

@hjonnala

we were able to resolve the issue by having a workaround with openvino2tensorflow :

Pytorch -> onnx -> openViNO -> edge_tpu

https://github.com/PINTO0309/openvino2tensorflow

blog: https://qiita.com/PINTO/items/ed06e03eb5c007c2e102

i try it but didnt work it gave me error
it give this error
line 520, in _quantize
return _mlir_quantize(
File "/usr/local/lib/python3.8/dist-packages/tensorflow/lite/python/convert_phase.py", line 218, in wrapper
raise error from None # Re-throws the exception.
File "/usr/local/lib/python3.8/dist-packages/tensorflow/lite/python/convert_phase.py", line 208, in wrapper
return func(*args, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/tensorflow/lite/python/convert.py", line 236, in mlir_quantize
return wrap_toco.wrapped_experimental_mlir_quantize(
File "/usr/local/lib/python3.8/dist-packages/tensorflow/lite/python/wrap_toco.py", line 47, in wrapped_experimental_mlir_quantize
return _pywrap_toco_api.ExperimentalMlirQuantizeModel(
RuntimeError: Failed to quantize: