stephantul 12 hours ago

Hello!

I recently did some experiments in decasing models. I noticed our users not using casing at all, while the base models we use tend to be cased. So I investigated, and found a nice way to decase transformers tokenizers and make them lowercase internally, which I call decasing. I wrote a blog about it!