I’ve been training (mostly finetuning) GPTs for a while and sometimes models turn out bad, even when loss/data seem fine. Is this normal? Should I expect to get a good model right away when I set up everything correctly, or is it up to chance?
I’ve been training (mostly finetuning) GPTs for a while and sometimes models turn out bad, even when loss/data seem fine. Is this normal? Should I expect to get a good model right away when I set up everything correctly, or is it up to chance?