search

antmicro-labs / voice-assistant

Apache License 2.0
0 stars 1 forks source link

Wierd tf behaviour. Probably out of memory on stack. #29

Closed szymon-filipiak closed 4 years ago

szymon-filipiak commented 4 years ago

We are observing weird behaviour when we load a larger tf model. We get false when matching the same values. If you have some time now we are working on it. If not, can you tell us how big can stack on our board be? Our model has approx 50kb @kgugala

zawadzki-dawid commented 4 years ago

We changed the model to the default one (yes, no) and we found out that voice recognition don't work. We tried to debug it but the problem seems to be outside of our purview. If you have time it would be great to schedule telco as soon as possible, even today, because we need to finish our project till friday.

rw1nkler commented 4 years ago

Yes, we can schedule the telco. @kgugala? The stack options can be found in the Zephyr Kernel Configuration (General Kernel Options submenu):

Selection

The configuration can be changed using the menuconfig target: https://docs.zephyrproject.org/latest/guides/kconfig/menuconfig.html#interactive-kconfig-interfaces

kgugala commented 4 years ago

I have a slot @2PM Is it OK for you to meet then?

rw1nkler commented 4 years ago

@szymon-filipiak @Marzecki @plonajakub @zawadzki-dawid @FilipKadziolka

szymon-filipiak commented 4 years ago

Yes, let's schedule a meeting.

szymon-filipiak commented 4 years ago

We have detected a mismatch between the new model's training scripts tf version and tf version in our code. There is a fine probability that retraining the net will solve the problem. @zawadzki-dawid is doing it right now.