Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
aari1995 
posted an update Feb 21
Post
looking at the tokenizer and the naming (“_en“), Google Gemma is very likely to have a multilingual counterpart. 👀

Thoughts?
deleted

No way it couldn't. The world has long since been globalized and there is always more money to be made in all languages than in just one.

If we got the model to use it's own internal language, and then translate it's interactions in english or whichever language, would be a better solution. Right now, adding too many different NLP languages risks bloating the model, so if google makes language specific models, that might be way too hard to do (and costly).

They use a subset of the Gemini tokenizer which is probably multilingual and perhaps multimodal