Sparkfun Edge Maximum Model Size?

Hello All,

I was wondering what is the maximum tflite model size that can be/has been used on the Sparkfun Edge/Edge 2?

The “audio classifier” seems to use 20kB of the 384kB RAM, but I’m wondering how much more there is to play with after the minimum system functionality has been bootloaded.

Thanks!

I’m also interested in this! Thank you for posting the question!

I am not familiar enough with TF to give a good answer off the top of my head. Instead I can suggest how to find out. You may compile the TF micro_speech example ([tutorial - note that paths have changed b/c the TF Lite ‘micro’ branch is no longer ‘experimental’). Then you may run the ‘size’ command (e.g. arm-none-eabi-size) on the resulting .elf or .axf file. That will show you the size of various sections. Here are some typically interesting sections:

.text - this is the amount of flash memory used to store the code itself

.data - this is the amount of RAM used by static / global variables (keep in mind that applications require more RAM than this number - some goes to allocation on the stack and some more may go to dynamic memory allocation if applicable)

So, when you run the ‘size’ command on the .elf or .axf file you should see that the .data section is at least 20 kB. (Unless TF stores the model in flash… I don’t know). Then you could figure out the size of statically allocated RAM usage that is not from the model by subtracting the size of the model. Something like:

S_non-model = S_data-section - S_model

Hope that gets you going in the right direction!](AI Speech Recognition with TensorFlow Lite for Microcontrollers and SparkFun Edge  |  Google Codelabs)