Fix couple of issues with inference, dataset folder

Add tokenized dataset, pre-training script

Add training and preprocessing scripts