Is distillation or synthetic data used during pre-training? If yes how much?