BertTokenizer not giving correct offset values for string having unicode characters

Question

I am working on the SQUAD 1.1 tfds dataset for a project that uses BERT. I needed offsets for each wordpiece token and thus decided to use the BertTokenizer class from tenso