Fine-tuning improves a language model's performance in a target domain but can weaken its general capabilities. Existing methods often face a trade-off betwe…
机构:上交
来源:arXiv 2610.41147 | AI4Papers 论文推荐平台