Why is Inference Time Different in ST Edge AI Cloud vs STM32Cube.AI Studio?

[Aqsa]

I am getting different inference times while running my machine learning model on the M55 board. On ST Edge AI Cloud, the inference time is 3.809 ms, but when I use STM32Cube.AI Studio, it shows 9.60941 ms.Can anyone help me understand why there is such a difference in the inference time?

[hamitiya]

Hello @Aqsa ,

Difference could be:

- Usage of binary mode for the weights

- Compilation flags for the ST Neural-ART Accelerator compiler (especially, `enable-epoch-controller`)

Clock should be the same if you work with STM32N6570-DK.

Could you please share with me the logs you have in STM32Cube AI Studio ?

Potentially, try ticking this box to see if it accelerates your inference time, and check the settings here:

in Memory Pool (under menu "Target")

in Profile settings, if you work with profile "Default"

Best regards,

Yanis