tokenized_dataset_train = LineByLineTextDataset(
    tokenizer= tokenizer,
    raw_datasets = dataset,
    max_length=256,
)