Convert tokens to strings before partial matching #194
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
BPE tokenizers encode whitespaces as a special character
Ġ
, and it is possible that another tokenizer may encode one or several strings as special character(s). Partial matching with the tokens directly thus prevents relevant matches from happening.We thus add a
convert_token_to_string
method that converts back these special character to the string they correspond to so we can partially match the corresponding tokens with the regex.The following snippet illustrates the problem and its solution:
Note: we could also iterate over the token ids and use
tokenizer.decode
. This may be simpler than adding a new function, especially since we wouldn't have to sort the vocabulary. But maybe less explicit as this may leave people wondering why we don't usevocabulary.keys()
directly.