Trustworthiness Costs of Domain Adaptation in Small Language Models:A Cross-Architecture Empirical Study
小语言模型领域适应的可信性代价:一项跨架构实证研究
专题命中 领域大模型 :language model(title,abstract);small language model(title,abstract);SLM(summary_cn,abstract);分类 cs.CL、cs.AI
AI总结 本文通过对三类SLM架构、三个领域、两类训练数据及四种微调策略的216组实验,量化了领域适应的可信性代价,发现对抗扰动训练数据可提升适应质量且不降低可信性,部分安全策略反而增加对抗伤害易感性。
Comments 13 pages, 7 tables, 2 appendices (Reproducibility Checklist; Software and Data Availability). Code, model checkpoints, and datasets publicly available at https://github.com/rbpdlf/slm-trw