重新审视大语言模型微调中的灾难性遗忘
Revisiting Catastrophic Forgetting in Large Language Model Tuning
AI总结：
本文首次揭示大语言模型损失景观平坦性与灾难性遗忘的直接关联，并提出用锐度感知最小化平坦化损失景观以缓解遗忘，实验证明其有效且能与现有抗遗忘策略互补。
摘要：
灾难性遗忘（CF）指模型在学习新数据时遗忘先前获得的知识。它损害了大语言模型（LLMs）在微调过程中的有效性，但其根本原因尚未得到彻底研究。本文迈出第一步，在大语言模型领域揭示了模型损失景观的平坦性与CF程度之间的直接联系。基于此，我们引入锐度感知最小化方法，通过平坦化损失景观来缓解CF。在三个广泛使用的微调数据集上、跨越不同模型规模的实验证明了我们的方法在缓解CF方面的有效性。分析表明，我们很好地补充了现有的抗遗忘策略，进一步增强了大语言模型对CF的抵抗力。
英文摘要：
Catastrophic Forgetting (CF) means models forgetting previously acquired knowledge when learning new data. It compromises the effectiveness of large language models (LLMs) during fine-tuning, yet the underlying causes have not been thoroughly investigated. This paper takes the first step to reveal the direct link between the flatness of the model loss landscape and the extent of CF in the field of LLMs. Based on this, we introduce the sharpness-aware minimization to mitigate CF by flattening the loss landscape. Experiments on three widely-used fine-tuning datasets, spanning different model scales, demonstrate the effectiveness of our method in alleviating CF. Analyses show that we nicely complement the existing anti-forgetting strategies, further enhancing the resistance of LLMs to CF.
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。
