In case you want all wordpieces of a word to be labeled, you can initialize the tokenizer with only_label_first_subword set to False.