That's a really nice figure! I would say in addition to all the parameter-efficient finetuning techniques, one category that is missing is full finetuning. That is, just training the model as is without PEFT.

→ View original post on X — @rasbt