在低秩适配器中高效实现类GEM持续学习，大幅降低计算开销。
GEM-Style Constraints for PEFT with Dual Gradient Projection in LoRA
- 在LoRA子空间内设计双梯度投影，约束适配器参数防止干扰
- 3个任务上精度接近原版GEM，投影耗时减少千倍以上
- 适合资源受限场景下的大模型持续学习应用
大语言模型全量微调计算成本高，促使采用参数高效适配器的持续学习方法。本文重新审视在低秩适配器（LoRA）子空间中应用梯度纪元记忆（GEM）机制，提出I-GEM：一种固定预算、驻留显存的双投影梯度近似方法，用于替代原GEM的二次投影。通过仅在适配器参数上施加非干扰约束，I-GEM在保持类GEM稳定性的同时，平均投影开销降低数个数量级。在带有领域偏移的3任务AG News划分上，使用GPT-2（355M）和LoRA（r=8），I-GEM的平均准确率与GEM相差仅约0.04个百分点，优于A-GEM约1.4个百分点。关键的是，其投影时间相比GEM减少约1000倍。结果表明，在LoRA子空间应用GEM约束是大规模大模型持续学习的可行路径。
原文摘要 · Abstract (English)
Full fine-tuning of Large Language Models (LLMs) is computationally costly, motivating Continual Learning (CL) approaches that utilize parameter-efficient adapters. We revisit Gradient Episodic Memory (GEM) within the Low-Rank Adapter (LoRA) subspace and introduce I-GEM: a fixed-budget, GPU-resident dual projected-gradient approximation to GEM's quadratic projection. By constraining non-interference solely within the adapter parameters, I-GEM preserves GEM-like stability with orders-of-magnitude lower mean projection overhead. On a 3-task AG News split with induced domain drift, using GPT-2 (355M) and LoRA ($r=8$), I-GEM matches GEM's average accuracy (within $\sim\!0.04$ pts) and outperforms A-GEM by $\sim\!1.4$ pts. Crucially, it reduces projection time vs.\ GEM by a factor of $\sim\!10^3$. These results suggest that applying GEM constraints in the LoRA subspace is a practical pathway for continual learning at the LLM scale.
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。