bert baseline pre and post process

--- dataset_info: features: - name: input_ids sequence: int32 - name: attention_mask sequence: int8 - name: special_tokens_mask sequence: int8 splits: - name: train num_byt