the size of samples in each k-fold on GPU is different from the size of samples in each k-fold on CPU

Question

I am running the same code training the same CNN model using the same dataset on GUP and CPU, and I am using k-fold cross validation in my code. The problem is that k-fold s