Didn't find op for builtin opcode 'CONV_2D' version '5' Failed to get registration from op code d

[Husnain-Bhatti]

@eloquentarduino I am trying to take an inference for a neural network on ESP 32 DEV KIT but it's throwing an error that is not understandable.

Please find my code below:

include

include <eloquent_tinyml/tensorflow.h>

include "test_model.h"

define NUMBER_OF_INPUTS 7200

define NUMBER_OF_OUTPUTS 4

define TENSOR_ARENA_SIZE 2*1024

Eloquent::TinyML::TensorFlow::TensorFlow<NUMBER_OF_INPUTS, NUMBER_OF_OUTPUTS, TENSOR_ARENA_SIZE> ml;

void setup() { Serial.begin(115200); ml.begin(test_model); }

void loop() { float x_test[7200] = {0.000000, 0.000139, ........ , 0.999722, 0.999861}; float y_pred[4] = {0, 0, 0, 0};

float predicted = ml.predict(x_test, y_pred);

Serial.print("\t predicted: ");
Serial.println(predicted);
delay(10000);

}

below is my model summary:

[eloquentarduino]

This kind of error is most times due to Tf version mismatch. Use Tf==2.2.4

[Husnain-Bhatti]

Thanks, @eloquentarduino for your response. This model is working on TensorFlow 2.1.1 or below versions. But softmax is not working on these versions. If I don't apply softmax on the last dense layer it works fine but when I place softmax on the dense layer it throws error

abort() was called at PC 0x400de461 on core 1

[koulnodji6056]

@Husnain-Bhatti please, have you been able to convert your model using tinymlgen after tfliteconverter?

[Husnain-Bhatti]

@koulnodji6056 yes I have been able to do so. The issue occurs when I run it on ESP32 DEV KIT.

[koulnodji6056]

@Husnain-Bhatti Hope someone with more experience will help with that. I have used LSTM and was not able ton convert it

[eloquentarduino]

Compatibility issues are pretty common with Tf, as you can see from other issues. I got replies from people who were able to deploy LSTM, so that's definetely possible. Please post the Python code so I can try to replicate.

[koulnodji6056]

Thank you for the quick replay. Here below is the link for the python code and the dataset. https://github.com/koulnodji6056/Prediction_LSTM_model_EloquentTinyML_library kindly let me know if you need any additional information

[koulnodji6056]

@eloquentarduino kindly share the link from the people who were able to deploy LSTM, so that we can follow as well. thank you

[eloquentarduino]

@koulnodji6056 I cannot find the links (if I remember correctly, nobody shared their code, they just said they got it working).