You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently we allow all encodings that satiisfy the constraints, but actually the tokenizer has a canonical encoding, which i generally the shortest and with priority.
For example,
"aaaa" can be tokenized as ["aa", "aa"] or ["aaa", "a"] or ["a", "aaa"] or more.
Currently our constraints allow all of them but actually only one is canonical.
One flag can be added to enfore strict mode
The text was updated successfully, but these errors were encountered:
Currently we allow all encodings that satiisfy the constraints, but actually the tokenizer has a canonical encoding, which i generally the shortest and with priority.
For example,
"aaaa" can be tokenized as ["aa", "aa"] or ["aaa", "a"] or ["a", "aaa"] or more.
Currently our constraints allow all of them but actually only one is canonical.
One flag can be added to enfore strict mode
The text was updated successfully, but these errors were encountered: