If you get a "missing tokenizer" error, run this line as well when importing the model:
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
