Not necessarily. Done right, full fine-tuning is usually better… my assumption for why this prevents overfitting is it’s not really affecting the parameters as much as a FT. Total speculation, I may be wrong.

→ View original post on X — @mattshumer_,