Spaces:

Xenova

/

the-tokenizer-playground

like 412

Running

The display of tokens based on color has a bug for complex scripts(Indic scripts, arabic etc).
For example,

I analysed the text "സാങ്കേതികമായി നന്ദി വിജയിച്ചെങ്കിലും സാമ്പത്തിക കാരണങ്ങളാൽ പദ്ധതി നിർത്തിവെച്ചു." using Gemma model. The content is in Malayalam - an Indic lanuage

As you can observe, the number of tokens rendered does not match the actual number of tokens. Let me provide a simplest version to understand this better

The input text is "സേ"

The two tokens are "സ", "േ". But together they form a ligature "സേ". Because of this, probably due to the bug in the color highlighting tool, only one color is used and that shows both the tokens in green color.