DP linear probing can outperform full fine-tuning, and a sequential LP-FT approach may achieve better test loss.
