Batching

Also written: batch, batch_size, batches

Running several inputs through a model in one call. Per-call overhead is shared and matrix math runs faster, so the cost per message drops.