Kernel Disconnects after a few epochs

I’m running Zero to JupyterHub on Kubernetes on Google Cloud. I have 8 GB ram (2vCPUs) machines. When my user runs a neural net (about 10k dataset) it fails after a few epochs and the kernel disconnects.
Increasing the RAM I think helps but I thought an 8 GB machine should be good enough. Is there any setting I have to set?