Who Learns Faster With Less Data? Adapters Beat Full Fine‑Tuning

Experiments show PEFT, especially Prefix‑tuning and UniPELT, outperform FFT in low‑resource text tasks and remain strong in AL setups, boosted further by TAPT.