Batch Size

How many training examples are processed together before the model’s weights are updated once. Larger batches use more VRAM but often train faster and more stably.