STABLE：面向大语言模型的门控持续学习
STABLE: Gated Continual Learning for Large Language Models
AI总结：
本文提出 STABLE 门控持续自编辑框架，利用 LoRA 与 EM、bits、KL 三类预算指标控制更新缩放或拒绝，在 Qwen-2.5-7B 上缓解灾难性遗忘并保持适应能力。
摘要：
大语言模型（LLM）日益需要在不进行全面重新训练的情况下实现持续适应的机制。然而，顺序更新可能导致灾难性遗忘，即新的编辑会降低先前获得的知识。本工作提出了 STABLE，这是一种门控持续自编辑框架，通过 Low Rank Adaptation（LoRA；见 arXiv:2106.09685）进行参数高效微调，从而在顺序更新期间约束遗忘。每个候选编辑都会依据稳定性预算，使用三种指标之一进行评估：(i) Exact Match（EM）下降，用于捕捉事实准确性损失；(ii) bits 增加，反映模型置信度降低；以及 (iii) KL 散度，用于量化基础模型与适应后模型之间的分布漂移。如果超过阈值，则通过裁剪过程重新缩放 LoRA 更新，或者拒绝该更新。在 Qwen-2.5-7B 模型上的实验表明，门控有效缓解了遗忘，同时保持了适应性。基于 EM 的门控在较短的持续学习序列中取得了最高的累积性能。我们的结果表明，不同的门控策略可以实现相当的分布偏移（以 KL 散度衡量），同时产生不同的准确性结果，这凸显了持续适应中门控设计的重要性。该方法为持续模型编辑提供了一种有原则的方法，使 LLM 能够整合新知识，同时保持可靠性。代码：https://github.com/Bhoy1/STABLE
英文摘要：
Large language models (LLMs) increasingly require mechanisms for continual adaptation without full retraining. However, sequential updates can lead to catastrophic forgetting, where new edits degrade previously acquired knowledge. This work presents STABLE, a gated continual self editing framework that constrains forgetting during sequential updates using parameter efficient fine tuning via Low Rank Adaptation (LoRA; see arXiv:2106.09685). Each candidate edit is evaluated against a stability budget using one of three metrics: (i) Exact Match (EM) drop, capturing factual accuracy loss; (ii) bits increase, reflecting reduced model confidence; and (iii) KL divergence, quantifying distributional drift between the base and adapted models. If a threshold is exceeded, the LoRA update is rescaled through a clipping procedure or rejected. Experiments on the Qwen-2.5-7B model show that gating effectively mitigates forgetting while preserving adaptability. EM based gating achieved the highest cumulative performance in short continual learning sequences. Our results show that different gating strategies can achieve comparable distribution shift (measured by KL divergence) while producing different accuracy outcomes, highlighting the importance of gating design in continual adaptation. This approach offers a principled method for continual model editing, enabling LLMs to integrate new knowledge while maintaining reliability. Code: https://github.com/Bhoy1/STABLE
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
