Jetson Nano won't train data in Juypter Labs


I am trying to follow the “Getting Started with AI on Jetson Nano” course and I am running into problems when I try to train data in Juypter Labs. The course says to take a couple of pictures with your thumbs up and then another couple of pictures with your thumbs down. After doing this you’re supposed to hit the train button and the Jetson is supposed to then start going through the data and learning what thumbs up and thumbs down is. Instead of doing this when I hit the train button nothing happens and then my jetson turns off a couple of minutes after. Does anyone know what the problem could be or what I am doing wrong? (Everything is on Linux)

Hi @patrickmarkoleary1, if your Nano abruptly powers off, typically this is related to a power supply issue. You may want to upgrade power supplies to one that can supply higher amperage.

If you run sudo nvpmodel -m 1, does it not turn off anymore? This will put the board in 5W mode, which draws less power (but also runs slower)

It is normal for PyTorch training to take a bit of time to start on the first epoch. It has to load a lot of code and CUDA kernels, some of which gets swapped out to your swapfile. You should check that you mounted at least 4GB of SWAP and disabled ZRAM per the DLI instructions.

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.