How to fine-tuning non-English data

[njawh]

Hello.
I would like to proceed with fine tuning with non-English speaking data.

I referred to the fine tuning guide you gave me before.(#70)
And I also referred to this person's post.(#157)

We found that the fine tuning code calls and uses the existing tokenizer, and the tokenizer does not update the new data.
Therefore, even if fine tuning is performed, non-English speaking data is provided with inconsistent outputs.

I would appreciate it if you could advise me on what to do.