PG-SFT: Balancing Capability Acquisition and Retention in Offline Agent Fine-Tuning
Read the original on arXiv AI →PG-SFT: Balancing Capability Acquisition and Retention in Offline Agent Fine-Tuning explores how to maintain a model’s existing abilities while teaching it new ones through supervised fine‑tuning on offline agent trajectories. The authors compare standard SFT, KL‑penalty, and update‑magnitude constraints, finding that these methods still degrade non‑target capabilities. They introduce Privilege‑Guided SFT (PG‑SFT), which uses turn‑level information gain to modulate supervision strength, achieving a better trade‑off between acquiring new skills and preserving existing ones, though with a slight drop in target‑task performance.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.