Zephon integrates with TorchTitan to preprocess training data on the fly and preserve identical global batches across GPU counts when the data recipe stays fixed.