Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2601.02500 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917185977319424 |
|---|---|
| author | Tekmen, Brian Yin, Jason Tong, Qianqian |
| author_facet | Tekmen, Brian Yin, Jason Tong, Qianqian |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_02500 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | GEM-Style Constraints for PEFT with Dual Gradient Projection in LoRA Tekmen, Brian Yin, Jason Tong, Qianqian Machine Learning Artificial Intelligence 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. |
| title | GEM-Style Constraints for PEFT with Dual Gradient Projection in LoRA |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2601.02500 |