Failed to allocate TFLite arena using XIAO ESP32S3

Help
#1

I trained a model and deployed it to XIAO ESP32S3.

RAM usage seems to be low according to this screenshot:

image
image795×472 51 KB

I created a file named boards.local.txt in C:\Users\username\AppData\Local\Arduino15\packages\esp32\hardware\esp32\2.0.9

esp32s3.build.extra_flags=-DEI_CLASSIFIER_ALLOCATION_STATIC

But the error persists.

I would appreciate tips on how to proceed. Thank you!

#2

The ESP S3 is a soon to be a fully supported board. But for now try:

#3

Thanks a lot! This resolved the error.

(@AIWintermuteAI Note that for I had to change resolver.AddUnidirectionalSequenceLstm(); to resolver.AddUnidirectionalSequenceLSTM();
in src\tflite-model\tflite-resolver.h to avoid a compilation error. I am using the newly available LSTM architecture)

1 Like
#4

Hi, @curious_cat !
I’ve merged fix to the issue you described. It should be deployed into production within next few days.
For full support for ESP32S3, there is a bit more work involved and I am on other task currently, so likely to be done in early October.

2 Likes
#5

Is there an approximate time, please? Indeed, replacing the ESP-NN is not a very convenient method.

#6

Perhaps as a temporary fix, ESP-NN could be replaced by default as part of the production code? although that would require specifying ESP32S3 before building and that might be more effort than it is worth