How dataset should i use to train a base model to teach an ability #738
Unanswered
fatihayUtopia
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello i have used "fatihay/TurkishTest" dataset to teach turkish to base model called "Qwen/Qwen2-1.5B-Instruct" but still he can not speak turkish as i expect. (in train process, last loss rate was 1.7) by the way i used autotrain with default settings.
1-Is my dataset form is right to train that base model
2-I will use my model as a customer representative, so my expectations from him are; When I give a system message containing my company's information, he should inform the user accordingly, follow the chat history closely, and not believe the user's misleading questions (for example, the user may say that the price of a product is more affordable than what is written in the system. The chatbot should remain loyal to the system message, not the user). Also. I want him to has the following ability: I will give him a json template as a system message and he will fill in the variables in the template by detecting it from the user's message and write the filled template as a response(for exaple: system: please fill the following template: {name surname: \n product code: }. user: hi what is the price of that red car the code was like t276. assistant: {name surname:none \n product code:t276 }). Can you suggest sample datasets? which datasets I can train with, according to my requests?
Beta Was this translation helpful? Give feedback.
All reactions