I’m working on a deep learning model with a training dataset containing 2500 observations. I’m uncertain about the optimal batch size to use for training to achieve the best balance between performance and efficiency.