Getting started building a new model #2
Description
Would you say a few more words about how to get started building a new model & what is necessary for such a task?
In particular, I'm wondering how to represent or build a tokenizer for a new language.
I'm also wondering, if I have gold NER, UPOS, and XPOS data, but no dependencies, is it possible to use those resources for the finetuning?
Also, I see some scripts referencing electra... is there a MicroElectra as part of this package as well?
Thanks in advance