Question During Training

I noticed that when I was training an existing dataset of 10k images (for training), only around 600 images were used for each epoch, while when I train one of my dataset (6120 images for training), it uses all 6120 images for one epoch.

  1. How do I make my own dataset to not use all images per each epoch?
  2. Will there be a decrease in accuracy by making it train on lower images per epoch?