已入深夜,您辛苦了!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您度过漫漫科研夜!祝你早点完成任务,早点休息,好梦!

Train Small, Infer Large: Memory-Efficient LoRA Training for Large Language Models

计算机科学 培训(气象学) 自然语言处理 人工智能 地理 气象学
作者
Jun Zhang,Jue Wang,Huan Li,Lidan Shou,Ke Chen,Yang You,Guiming Xie,Xuejian Gong,Kaijun Zhou
出处
期刊:Cornell University - arXiv
标识
DOI:10.48550/arxiv.2502.13533
摘要

Large Language Models (LLMs) have significantly advanced natural language processing with exceptional task generalization capabilities. Low-Rank Adaption (LoRA) offers a cost-effective fine-tuning solution, freezing the original model parameters and training only lightweight, low-rank adapter matrices. However, the memory footprint of LoRA is largely dominated by the original model parameters. To mitigate this, we propose LoRAM, a memory-efficient LoRA training scheme founded on the intuition that many neurons in over-parameterized LLMs have low training utility but are essential for inference. LoRAM presents a unique twist: it trains on a pruned (small) model to obtain pruned low-rank matrices, which are then recovered and utilized with the original (large) model for inference. Additionally, minimal-cost continual pre-training, performed by the model publishers in advance, aligns the knowledge discrepancy between pruned and original models. Our extensive experiments demonstrate the efficacy of LoRAM across various pruning strategies and downstream tasks. For a model with 70 billion parameters, LoRAM enables training on a GPU with only 20G HBM, replacing an A100-80G GPU for LoRA training and 15 GPUs for full fine-tuning. Specifically, QLoRAM implemented by structured pruning combined with 4-bit quantization, for LLaMA-3.1-70B (LLaMA-2-70B), reduces the parameter storage cost that dominates the memory usage in low-rank matrix training by 15.81$\times$ (16.95$\times$), while achieving dominant performance gains over both the original LLaMA-3.1-70B (LLaMA-2-70B) and LoRA-trained LLaMA-3.1-8B (LLaMA-2-13B).
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
稻槐发布了新的文献求助10
1秒前
科研通AI6.3应助柏1Y采纳,获得10
1秒前
汉堡包应助LSH970829采纳,获得10
1秒前
2秒前
Jrayty发布了新的文献求助10
2秒前
4秒前
拂晓神剑完成签到,获得积分10
4秒前
仗炮由纪发布了新的文献求助10
5秒前
糖炒板栗完成签到,获得积分10
5秒前
科研通AI6.4应助wang采纳,获得10
5秒前
超级访冬发布了新的文献求助10
6秒前
skopy发布了新的文献求助10
7秒前
8秒前
9秒前
9秒前
千束完成签到,获得积分10
12秒前
popcorn发布了新的文献求助10
14秒前
研友_VZG7GZ应助核桃采纳,获得10
14秒前
15秒前
15秒前
WW发布了新的文献求助10
16秒前
深情安青应助JT采纳,获得10
16秒前
17秒前
17秒前
无极微光应助千束采纳,获得20
18秒前
19秒前
核桃应助文件撤销了驳回
20秒前
隐形曼青应助WW采纳,获得10
21秒前
shjyang完成签到,获得积分0
21秒前
FashionBoy应助阳光的白晴采纳,获得10
22秒前
22秒前
劳永杰发布了新的文献求助10
23秒前
顾旻发布了新的文献求助30
27秒前
Jacquielin完成签到 ,获得积分10
27秒前
32秒前
33秒前
33秒前
我是老大应助劳永杰采纳,获得10
33秒前
oqo完成签到 ,获得积分10
37秒前
ddl完成签到,获得积分10
39秒前
高分求助中
Invited Discussant 63O and 64O 1000
Ideology and Meaning-Making under the Putin Regime 750
Petrology and Plate Tectonics 500
A Handbook of User Experience Research & Design in Libraries 400
Understanding Modeling and Simulation of Polymerization Reactions 400
Direct and Iterative Linear System Solvers 400
《KNN基无铅压电陶瓷电学性能优化与物理机理研究》 300
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 工程类 有机化学 计算机科学 化学工程 生物化学 物理 内科学 复合材料 催化作用 光电子学 物理化学 电极 细胞生物学 基因 遗传学
热门帖子
关注 科研通微信公众号,转发送积分 6907051
求助须知:如何正确求助?哪些是违规求助? 8600325
关于积分的说明 18256017
捐赠科研通 6311939
什么是DOI,文献DOI怎么找? 3064631
关于科研通互助平台的介绍 2088158
邀请新用户注册赠送积分活动 2042295