TFLite mismatch: Didn't find op for builtin opcode 'CONV_2D' version '3'

[ksaye]

Following https://github.com/pschatzmann/arduino-audio-tools/wiki/TensorFlow-Lite---MicroSpeech, I was able to generate the model file: https://kevinsayazstorage.blob.core.windows.net/public/model.h.

Using the sample stream-i2s-tf works great with the included model.h.

When I add my model and make the following change (I changed the WANTED_WORDS = "marvin,stop,off,on,go,sheila") in the notebook:
const char* kCategoryLabels[8] = {"silence","unknown","marvin","stop","off","on","go","sheila"};

I get an error:

  Didn't find op for builtin opcode 'CONV_2D' version '3'. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?
  
  Failed to get registration from op code CONV_2D
   
  [E] TfLiteAudioStream.h : 857 - AllocateTensors() failed
  Guru Meditation Error: Core  1 panic'ed (StoreProhibited). Exception was unhandled.
  Core 1 register dump:
  PC      : 0x4000c2e4  PS      : 0x00060b30  A0      : 0x800d31d7  A1      : 0x3ffb1eb0  
  A2      : 0x00000000  A3      : 0x3ffd03a0  A4      : 0x000007a8  A5      : 0x00000000  
  A6      : 0x00000000  A7      : 0x00000000  A8      : 0x00000000  A9      : 0x000000b1  
  A10     : 0x0001cd4d  A11     : 0x00000028  A12     : 0x3ffbc018  A13     : 0x00000016  
  A14     : 0x3ffbc068  A15     : 0x00000000  SAR     : 0x0000001a  EXCCAUSE: 0x0000001d  
  EXCVADDR: 0x00000000  LBEG    : 0x4000c2e0  LEND    : 0x4000c2f6  LCOUNT  : 0x00000079  
  
  ELF file SHA256: 0000000000000000
  
  Backtrace: 0x4000c2e4:0x3ffb1eb0 0x400d31d4:0x3ffb1ec0 0x400d2d8e:0x3ffb1ef0 0x400d2e5d:0x3ffb1f10 0x400d4b82:0x3ffb1f30 0x400d4d1b:0x3ffb1f50 0x400d4d86:0x3ffb1f90 0x40105b79:0x3ffb1fb0 0x4008649e:0x3ffb1fd0

  Rebooting...

Is this a known issue?

[pschatzmann]

My first guess would be that it is running out of RAM. You would need to analyze the stacktrace to confirm this.
If you have PSRAM you could try to activate the use of PSRAM
https://www.pschatzmann.ch/home/2022/05/30/esp32-and-psram/

[ksaye]

I just recompiled the model, this time with fewer words. The sample model.h you include [char array] is of size: 18715. My model is of size 14752, which is smaller and I get the same error.

Are you sure there has not been an update in the model? I see a cell in the notebook runs: git clone -q --depth 1 https://github.com/tensorflow/tensorflow and I am wondering if something in the tenforflow repo has changed.

[ksaye]

I also get the same error with: https://github.com/pschatzmann/tflite-micro-arduino-examples/blob/main/examples/micro_speech/micro_speech.ino, once I replace the model.cc and adjust the kCategoryLabels.

setup - begin
Didn't find op for builtin opcode 'CONV_2D' version '3'. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?

Failed to get registration from op code CONV_2D
 
AllocateTensors() failed
Guru Meditation Error: Core  1 panic'ed (LoadProhibited). Exception was unhandled.
Core 1 register dump:
PC      : 0x400d32c3  PS      : 0x00060130  A0      : 0x800d3524  A1      : 0x3ffb1f50  
A2      : 0x00000000  A3      : 0x3ffbdbcc  A4      : 0x00000000  A5      : 0x00000028  
A6      : 0x3ffb1f88  A7      : 0x3ffc07b8  A8      : 0x00000002  A9      : 0x3ffb1f40  
A10     : 0x3ffba4e4  A11     : 0x00000000  A12     : 0x00000029  A13     : 0x0000efc0  
A14     : 0x00000008  A15     : 0x3f400e94  SAR     : 0x0000000a  EXCCAUSE: 0x0000001c  
EXCVADDR: 0x00000000  LBEG    : 0x40088811  LEND    : 0x40088821  LCOUNT  : 0xfffffff9  

ELF file SHA256: 0000000000000000

Backtrace: 0x400d32c3:0x3ffb1f50 0x400d3521:0x3ffb1f80 0x400df78d:0x3ffb1fb0 0x4008a33e:0x3ffb1fd0

Rebooting...

[ksaye]

@pschatzmann

I have spent days trying to resolve this. Using your model.h it works. When I compile my own, I get the error below. I have tried a different tag version of tensorflow, nothing works.

I still get:

16:04:48.365 -> Didn't find op for builtin opcode 'CONV_2D' version '3'. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?
16:04:48.397 -> 
16:04:48.397 -> Failed to get registration from op code CONV_2D
16:04:48.397 ->  
16:04:48.397 -> [E] TfLiteAudioStream.h : 857 - AllocateTensors() failed
16:04:48.397 -> Guru Meditation Error: Core  1 panic'ed (StoreProhibited). Exception was unhandled.
16:04:48.429 -> Core 1 register dump:
16:04:48.429 -> PC      : 0x4000c2e4  PS      : 0x00060b30  A0      : 0x800d31d7  A1      : 0x3ffb1eb0  
16:04:48.429 -> A2      : 0x00000000  A3      : 0x3ffc7c58  A4      : 0x000007a8  A5      : 0x00000000  
16:04:48.429 -> A6      : 0x00000000  A7      : 0x00000000  A8      : 0x00000000  A9      : 0x000000b3  
16:04:48.429 -> A10     : 0x0001d24d  A11     : 0x00000028  A12     : 0x3ffbab90  A13     : 0x00000016  
16:04:48.472 -> A14     : 0x3ffbabe0  A15     : 0x00000000  SAR     : 0x0000001a  EXCCAUSE: 0x0000001d  
16:04:48.472 -> EXCVADDR: 0x00000000  LBEG    : 0x4000c2e0  LEND    : 0x4000c2f6  LCOUNT  : 0x00000079  
16:04:48.472 -> 
16:04:48.472 -> ELF file SHA256: 0000000000000000
16:04:48.472 -> 
16:04:48.472 -> Backtrace: 0x4000c2e4:0x3ffb1eb0 0x400d31d4:0x3ffb1ec0 0x400d2d8e:0x3ffb1ef0 0x400d2e5d:0x3ffb1f10 0x400d4b82:0x3ffb1f30 0x400d4d1b:0x3ffb1f50 0x400d4d86:0x3ffb1f90 0x40105b79:0x3ffb1fb0 0x4008649e:0x3ffb1fd0
16:04:48.502 -> 
16:04:48.502 -> Rebooting...

When I try to run: https://github.com/pschatzmann/arduino-audio-tools/tree/main/examples/examples-stream/streams-i2s-tf

As you can see from: fail_train_micro_speech_model.zip, I can train the model but when I copy model.cc it compiles but errors as above. Can you take a quick look? model.zip

[pschatzmann]

When I created the library it was for version=2.4.0-ALPHA
Did you try to make sure that you used the correspinding version in Jupyter ?

[ksaye]

I have tried a few version, but not that version. Let me git it a whirl.

[ksaye]

I tried in cell 5 to use !git clone -q --depth 1 https://github.com/tensorflow/tensorflow --branch v2.4.0 and !git clone -q --depth 1 https://github.com/tensorflow/tensorflow --branch v2.3.0, neither worked.