CUDA out of memory when running Bert with Pytorch (Previously worked)

Question

I am building a BERT binary classification on SageMaker using Pytorch.

Previously when I ran the model, I set the Batch size to 16 and the model were able to run successf