ChatGPT and by extension the GPT-4 API doesn’t work very well in my native language (it does understand it to some degree but not great.) If I had a large corpus of data would it be possible for me to fine-tune the standard model to better understand? Or would that require completely retraining an LLM?
2 Likes
Hey Chezky,
OFF Topic: I’m trying to learn how to create a custom GPT model for Yiddish. Care to share your knowledge or resources?
Thanks
1 Like
If you put Yiddish dictionary and thesaurus in its knowledge and tell it to reference it and use it to structure responses for questions regarding Yiddish education or language structure in its instructions . It may work, I use English thesaurus to add neuance to RPG GPT I build. Useing same logic.
This thing with Yiddish is that there are two dialect. One is not really is use but it is more academic, so most of the subject matter material and translators use that. The other one is wildly in use among native speakers.
There are a couple Yiddish language forums online which I want to use to train the model. How do I got about doing it?
1 Like
Find a slang pdf. Or a open source compilation.
Ask the forum for help make it a forum project.
I do this with RPGs whole FB groups hive mind.
You could also write your own.
Copy paste and a cloud notebook is your friend.
