fine-tuning and model capability

#2
by vtiyyal1 - opened

I tried the online chat llama.family it's generating english sentences as a reply even after modifying the system prompt. Is this because the model is not really aligned for QA task? because of the architecture ?
I am asking this because I fine-tuned on QA task but with very lengthy contexts, but looks like the model is not learning much, it doesn't generate anything when asked something in lengthy after fine-tuning. It generates responses to small length questions. I'm confused if whether I did something wrong while fine-tuning or it's just the model architecture. Also I would like to know more about your approach of training it for QA task for translation. Would appreciate any suggestions.

Sign up or log in to comment