Understanding the output fields of Profile tool

Help
#1

Question/Issue:
I want to understand the output of model.profile()
Assume I have something like this :

{
“device”: “alif-he”,
“tfliteFileSizeBytes”: 135816,
“isSupportedOnMcu”: true,
“memory”: {
“tflite”: {
“ram”: 391896,
“rom”: 191976,
“arenaSize”: 391552
},
“eon”: {
“ram”: 322720,
“rom”: 150632
}
},
“timePerInferenceMs”: 2527
}

Are ‘ram’ and ‘arenaSize’ additive. From my understanding, arenaSize would reside in RAM. Does it mean that the total ‘ram’ requriement is 391896 + 391552 or does it mean the total RAM requirement is just 391896, of which, 391552 is consumed by tensor_arena?

Project ID:

Context/Use case:
Model RAM usage profiling

1 Like
#2

Hi @tkaushik

Thank you for asking this, I can see how it could be misleading, your later guess is correct : 391896 is the total

i.e.

The arenaSize parameter represents the size of the tensor arena allocated within the total listed in ram.

I will make an update to the docs to clarify this.

Best

Eoin

1 Like