I Trained Llama3.2 using QLORA for Question Answering Task but model is not performing as expected

[Aravinth]

I prepare a custom dataset contains three columns (Context, Question, Answer) and train the model using Hugging Face SFT Trainer.

But the model generated output with out producing the end token. The output also not impressive. Can i know the reason from anyone. I have limited computer resource of 24 GB VRAM only so please let me know any better solution