AI Blog

AI Blog

by Michele Laurelli

Batch Size

/bætʃ saɪz/
Training
Definition

The number of training examples used in one iteration of model training.

Batch size affects training speed, memory usage, and model convergence. Small batches provide noisier gradients, large batches are more stable but memory-intensive. Common values: 32, 64, 128, 256.

Examples

1

Batch size 32 for limited GPU memory

2

Batch size 256 for faster training

3

Mini-batch gradient descent

Michele Laurelli - AI Research & Engineering