Currently, tok2spans.iob2spans accepts parallel lists of tokens and IOB-style labels. Since there is no single text, it constructs that text by concatenating the tokens with a single space as a delimiter.
It would be nice to support more flexible tokenization. One possibility is to replace the list of tokens with an existing text together with a tokenization of that text (including mapping tokens to spans).
Currently, tok2spans.iob2spans accepts parallel lists of tokens and IOB-style labels. Since there is no single text, it constructs that text by concatenating the tokens with a single space as a delimiter.
It would be nice to support more flexible tokenization. One possibility is to replace the list of tokens with an existing text together with a tokenization of that text (including mapping tokens to spans).