there is the word "Assistant" at the beginning of the fine tuning model response when in llama-cpp | Unsloth AI | Page 1

outer summit Nov 30, 2024, 9:11 AM

#

hi

clear snow Nov 30, 2024, 9:11 AM

#

hello

outer summit Nov 30, 2024, 9:18 AM

#

hi

outer summit Nov 30, 2024, 9:20 AM

#

clear snow hello

did you solve problem?

clear snow Nov 30, 2024, 9:21 AM

#

outer summit did you solve problem?

yes

outer summit Nov 30, 2024, 9:21 AM

#

great

#

could you tell me how to solve?

clear snow Nov 30, 2024, 9:23 AM

#

i change instruction system

from "You are an experienced customer support person named Tono.\nBe polite to users and answer all Pegon related queries."
to "Be polite to users and answer all Pegon related queries."

there is still the word "assistant\n" but it is clear and not messy anymore so I just delete it with output_text.replace("assistant\n", "")

outer summit Nov 30, 2024, 9:24 AM

#

nice

#

but I think when you fine tune the model, you maybe used another format

#

or how about change the prompt to remove that?

#

after changing prompt, if there it is still, the error is not in model. just in output format

#

how about your opinion?

#there is the word "Assistant" at the beginning of the fine tuning model response when in llama-cpp