pub struct Fold;Expand description
The default marker removal and Unicode comparison form.
The fixed control tokens [CLS], [SEP], [PAD], [UNK], [MASK],
<s>, </s>, <pad>, <unk>, <mask>, <cls>, <sep>,
<|endoftext|>, and <|startoftext|> produce an empty comparison form.
One leading ## and one trailing </w> are removed. Every ▁, Ġ, and
Ċ is dropped. Ġ and Ċ are also Maltese capital letters, so literal
Maltese text containing either character loses it under this comparison
form. Use a custom Normalizer when those letters carry text.
Each remaining character receives recursive compatibility decomposition
from the Unicode Character Database 17.0.0 data. The decomposition is
canonically ordered within that character, then mapped with the simple
lowercase field. Final sigma ς maps to sigma σ. Characters are folded
independently, so marks are not reordered across input characters.