[behnamoh]

but doesn’t finetuning result in forgetting previous knowledge? it seems that finetuning is most usable to train “structures” not new knowledge. am i missing something?