I am trying to train a Convolution Neural Network, but as the code starts executing, at some point on the right top corner it shows me a message showing that I have low memory. What is this? What should I do? I use 128GB mircoSD which is pretty enough I think for the work I am trying to accomplish. And jetson nano has 4 GB, I cannot understand where is the problem…
Jetson platform is designed for the run-time inference, which is not suitable for the training job.
Nano just has 4GB memory and need to share by the whole system.
If you want to apply a training on Nano, please reduce the input batchsize or use a much light-weight model.
I followd this video: https://www.youtube.com/watch?v=JXv39FGi-nw which works! But how can I reduce the Swap memory? For example if I had chosen 6GB swap memory and I want to reduce it to 3GB swap memory? because when I try the command:
./installSwapfile.sh -s 3GB
I get the response:
fallocate: invalid length value specified
You will need to recreate the swapfile first.
$ sudo swapoff swapfile #unmount swapfile $ sudo remove swapfile #remove the 6GB swapfile $ ./installSwapfile.sh -s 3GB #recreate a new one