arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2502.07752cs.LGcs.AIstat.ML
收藏

通过带低秩扩展的结构化Fisher近似实现大语言模型高效优化器设计

Towards Efficient Optimizer Design for LLM via Structured Fisher Approximation with a Low-Rank Extension

Microsoft Research(微软研究院)

Wenbo Gong, Meyer Scetbon, Chao Ma, Edward Meeds

AI总结:

本研究从结构化Fisher信息矩阵近似视角出发,提出两项大语言模型高效优化器设计策略,推导了RACS和Alice两种新型低内存优化器,在LLaMA预训练实验中实现了优于现有基线的收敛速度与内存效率。

摘要:

为大语言模型(LLM)设计兼具低内存需求与快速收敛性的高效优化器是一项重要且具有挑战性的问题。本文通过结构化Fisher信息矩阵(FIM)近似的视角,为此类优化器的系统化设计迈出了一步。我们证明,许多当前最先进的高效优化器都可被视为在特定结构假设下(基于Frobenius范数)的FIM近似解。基于这些见解,我们针对LLM提出了两项实用高效优化器的设计建议,包括精心选择结构假设以平衡通用性与效率,以及通过一种新颖的低秩扩展框架提升具有通用结构的优化器的内存效率。我们通过推导两种新型内存高效优化器——行与列缩放随机梯度下降(Row and Column Scaled SGD, RACS)和自适应低维子空间估计(Adaptive low-dimensional subspace estimation, Alice),展示了每种设计方法的应用方式。在LLaMA预训练(最高10亿参数规模)上的实验验证了方法的有效性,结果表明其收敛速度和效果均优于现有内存高效基线优化器以及内存开销极小的Adam。值得注意的是,Alice的收敛速度比Adam快2倍以上,而RACS在10亿参数模型上以接近SGD的内存占用实现了优异性能。

英文摘要:

Designing efficient optimizers for large language models (LLMs) with low-memory requirements and fast convergence is an important and challenging problem. This paper makes a step towards the systematic design of such optimizers through the lens of structured Fisher information matrix (FIM) approximation. We show that many state-of-the-art efficient optimizers can be viewed as solutions to FIM approximation (under the Frobenius norm) with specific structural assumptions. Building on these insights, we propose two design recommendations of practical efficient optimizers for LLMs, involving the careful selection of structural assumptions to balance generality and efficiency, and enhancing memory efficiency of optimizers with general structures through a novel low-rank extension framework. We demonstrate how to use each design approach by deriving new memory-efficient optimizers: Row and Column Scaled SGD (RACS) and Adaptive low-dimensional subspace estimation (Alice). Experiments on LLaMA pre-training (up to 1B parameters) validate the effectiveness, showing faster and better convergence than existing memory-efficient baselines and Adam with little memory overhead. Notably, Alice achieves better than 2x faster convergence over Adam, while RACS delivers strong performance on the 1B model with SGD-like memory.

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。

↑