arXiv 2308.087472025-01-05cs.CL

大型语言模型在持续微调过程中灾难性遗忘的实证研究

An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning

Pattern Recognition Center, WeChat AI, Tencent Inc（腾讯公司微信AI模式识别中心）；School of Engineering, Westlake University（西湖大学工程学院）；Institute of Advanced Technology, Westlake Institute for Advanced Study（西湖高等研究院先进技术研究所）

Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, Yue Zhang

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文实证研究 1b–7b 大型语言模型在持续指令微调中的灾难性遗忘，发现更大模型遗忘更严重、BLOOMZ 优于 mT0，且通用指令微调可缓解遗忘。

摘要：

灾难性遗忘（CF）是机器学习中的一种现象：模型在获取新知识以在下游任务中取得令人满意的性能时，会遗忘先前学习到的信息。随着大型语言模型（LLM）展现出卓越性能，研究 LLM 在持续指令微调过程中是否存在 CF 颇具意义。本研究从领域知识、推理和阅读理解的角度，实证评估了 LLM 在持续指令微调过程中的知识遗忘现象。实验表明，在参数规模从 1b 到 7b 的 LLM 中普遍观察到灾难性遗忘。令人惊讶的是，在该模型规模范围内，随着模型规模增大，遗忘严重程度反而加剧，这可能源于较大 LLM 的初始性能显著更高。将 decoder-only 模型 BLOOMZ 与 encoder-decoder 模型 mT0 进行比较，BLOOMZ 表现出更少的遗忘并保留了更多知识。有趣的是，我们还观察到 LLM 在持续微调过程中能够缓解性别偏见等语言偏见。此外，研究结果表明，通用指令微调有助于缓解 LLM 在后续微调过程中的遗忘现象。

英文摘要：

Catastrophic forgetting (CF) is a phenomenon that occurs in machine learning when a model forgets previously learned information while acquiring new knowledge for achieving a satisfactory performance in downstream tasks. As large language models (LLMs) have demonstrated remarkable performance, it is intriguing to investigate whether CF exists during the continual instruction tuning of LLMs. This study empirically evaluates the forgetting phenomenon in LLMs' knowledge during continual instruction tuning from the perspectives of domain knowledge, reasoning, and reading comprehension. The experiments reveal that catastrophic forgetting is generally observed in LLMs ranging from 1b to 7b parameters. Surprisingly, as the model scale increases, the severity of forgetting intensifies in such a model sale range which may result from the much significant initial performance in the larger LLM. Comparing the decoder-only model BLOOMZ with the encoder-decoder model mT0, BLOOMZ exhibits less forgetting and retains more knowledge. Interestingly, we also observe that LLMs can mitigate language biases, such as gender bias, during continual fine-tuning. Furthermore, our findings indicate that general instruction tuning can help alleviate the forgetting phenomenon in LLMs during subsequent fine-tuning.

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。