Didn't find op for builtin opcode 'SOFTMAX' version '2'. An older version of this builtin might be supported. Are you using an old TFLite binary with a newer model?

[AdamMiltonBarker]

Arduino BLE Sense
Tensorflow 2.5.0

I have a model that is 179960 len, had quite a lot of issues with the arena maxing out memory when it gets to the interpreter. It was requiring ~627 x 1024 but that was crashing memory. I now have the arena set at 200 x 1024 and using micro op resolver, and suddenly it started working, then I was hit with this.

Didn't find op for builtin opcode 'SOFTMAX' version '2'

I know it can be solved with AllOpsResolver but then throws the original issue:

Needed 627488 but only 203072 was available.

If I increase to 627 x 1024 it won't even upload.

Can anyone suggest how can be solved without having to use AllOpsResolver, haven't been able to find this issue documented anywhere. Thanks in advance.

[advaitjain]

Can you make sure that all the ops in your model have been explicitly added

tflite-micro/tensorflow/lite/micro/micro_mutable_op_resolver.h

Line 455 in 2ddb3cc

TfLiteStatus AddSoftmax(

Here's an example from the keyword benchmark

tflite-micro/tensorflow/lite/micro/benchmarks/keyword_benchmark.cc

Lines 54 to 58 in 2ddb3cc

	KeywordOpResolver* op_resolver = new (op_resolver_buffer) KeywordOpResolver();
	op_resolver->AddFullyConnected(tflite::Register_FULLY_CONNECTED_INT8());
	op_resolver->AddQuantize();
	op_resolver->AddSoftmax(tflite::Register_SOFTMAX_INT8_INT16());
	op_resolver->AddSvdf();

[AdamMiltonBarker]

Thank you for quick reply. These are the ops:

static tflite::MicroMutableOpResolver<8> micro_op_resolver;
micro_op_resolver.AddPad();
micro_op_resolver.AddConv2D();
micro_op_resolver.AddPad();
micro_op_resolver.AddConv2D();
micro_op_resolver.AddMaxPool2D();
micro_op_resolver.AddReshape();
micro_op_resolver.AddFullyConnected();
micro_op_resolver.AddSoftmax();

I will check out the links you gave, thanks buddy.

[AdamMiltonBarker]

OK I now understand more about ops in tflite and where I made my mistake. I made a new model, this has actually been a blessing in disguise. The model is uploading fine to Arduino, and the model is pretty damn good. Sorry for the noob questions but you helped me understand more about micro models :)