通过参数对齐缓解多语言模型持续训练中的灾难性遗忘问题
Parameter Alignment Mitigates Catastrophic Forgetting in Multilingual Expert Language Models
- 设计五种分层参数对齐策略，防止多语言持续预训练时参数漂移
- 在32种语言上测试，显著减少遗忘，同时保持新语言学习能力
- 不同策略适配不同任务：理解用正则化，翻译用回滚法
持续预训练（CPT）是将大语言模型扩展到新语言的实用方法，但直接在目标语言数据上微调会导致已有能力因灾难性遗忘而退化。按语系组织训练可减少跨语言干扰，但无法阻止通用知识遗忘。我们发现遗忘源于多语言CPT中的参数漂移，并提出五种分层参数对齐策略：硬层冻结、软正则化、事后权重回滚与模型融合。我们在涵盖五个语系共32种训练语言及未见语言的基准上，从困惑度、阅读理解、物理推理和翻译四个维度系统评估这些策略，对比两种无正则化的CPT基线。参数对齐显著降低遗忘，代价极小：层冻结与正则化最有效提升理解能力，事后回滚在翻译上表现最佳。结果揭示了家族专家型CPT的习得-遗忘边界，为不同任务匹配最优策略提供实用指导。
原文摘要 · Abstract (English)
While continual pretraining~(CPT) is a practical way to extend large language models to new languages, naïve finetuning on targeted data erodes existing capabilities through catastrophic forgetting. Organizing training around language families reduces cross-language interference but cannot alone prevent forgetting of the general knowledge needed for downstream tasks. We link this forgetting to parameter drift in multilingual CPT and present a suite of five layer-aware parameter alignment strategies: hard layer freezing, soft regularization, post-hoc weight reversion, and model merging. We systematically compare our alignment strategies against two unregularized CPT baselines on benchmarks spanning 32 training languages from five language families, plus held-out languages, across four evaluation axes: perplexity, reading comprehension, physical reasoning, and translation. Parameter alignment substantially reduces forgetting at minimal cost to language acquisition: layer freezing and regularization best preserve comprehension, whereas post-hoc reversion yields the strongest translation gains. Together, these results map the acquisition--forgetting frontier for family-expert CPT and offer practical deployment guidelines pairing each strategy to the tasks it best serves.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。