Scale-Distribution Decoupling: Enabling Stable and Effective Training of Large Language Models

• URL: http://arxiv.org/abs/2502.15499v2
• Date: Tue, 25 Feb 2025 06:54:39 GMT
• Title: Scale-Distribution Decoupling: Enabling Stable and Effective Training of Large Language Models
• Authors: Ya Wang, Zhijian Zhuo, Yutao Zeng, Xun Zhou, Jian Yang, Xiaoqing Li,
• Abstract summary: Training stability is a persistent challenge in the pre-training of large language models (LLMs)<n>We propose Scale-Distribution Decoupling (SDD), a novel approach that stabilizes training by explicitly decoupling the scale and distribution of the weight matrix in fully-connected layers.
• Score: 21.16132396642158
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Training stability is a persistent challenge in the pre-training of large language models (LLMs), particularly for architectures such as Post-Norm Transformers, which are prone to gradient explosion and dissipation. In this paper, we propose Scale-Distribution Decoupling (SDD), a novel approach that stabilizes training by explicitly decoupling the scale and distribution of the weight matrix in fully-connected layers. SDD applies a normalization mechanism to regulate activations and a learnable scaling vector to maintain well-conditioned gradients, effectively preventing $\textbf{gradient explosion and dissipation}$. This separation improves optimization efficiency, particularly in deep networks, by ensuring stable gradient propagation. Experimental results demonstrate that our method stabilizes training across various LLM architectures and outperforms existing techniques in different normalization configurations. Furthermore, the proposed method is lightweight and compatible with existing frameworks, making it a practical solution for stabilizing LLM training. Code is available at https://github.com/kaihemo/SDD.

Related papers

• Optimizing ML Training with Metagradient Descent [69.89631748402377]
  We introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a "smooth model training" framework that enables effective optimization using metagradients.
  arXiv  Detail & Related papers  (2025-03-17T22:18:24Z)
• LORENZA: Enhancing Generalization in Low-Rank Gradient LLM Training via Efficient Zeroth-Order Adaptive SAM [13.180761892449736]
  We study robust parameter-efficient fine-tuning (PEFT) techniques for large-language models (LLMs) We present a new highly computationally efficient framework called AdaZo-SAM, combining Adam and Sharpness-Aware Minimization (SAM) We also design a low-rank gradient optimization method named LORENZA, which is a memory-efficient version of AdaZo-SAM.
  arXiv  Detail & Related papers  (2025-02-26T21:30:34Z)
• RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models [95.32315448601241]
  We propose an algorithm named Rotated Straight-Through-Estimator (RoSTE)<n>RoSTE combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy to reduce activation outliers.<n>Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration.
  arXiv  Detail & Related papers  (2025-02-13T06:44:33Z)
• EDiT: A Local-SGD-Based Efficient Distributed Training Method for Large Language Models [4.514681046629978]
  We propose EDiT, an innovative Efficient Distributed Training method that combines a tailored Local SGD approach with model sharding techniques to enhance large-scale training efficiency.<n>We also introduce A-EDiT, a fully asynchronous variant of EDiT that accommodates heterogeneous clusters.<n> Experimental results demonstrate the superior performance of EDiT/A-EDiT, establishing them as robust solutions for distributed LLM training.
  arXiv  Detail & Related papers  (2024-12-10T06:08:24Z)
• AdaRankGrad: Adaptive Gradient-Rank and Moments for Memory-Efficient LLMs Training and Fine-Tuning [9.51289606759621]
  Training and fine-tuning large language models (LLMs) come with challenges related to memory and computational requirements.<n>Various techniques have been developed to tackle these challenges, such as low-rank adaptation (LoRA)<n>We introduce a new method inspired by a phenomenon we formally prove: as training progresses, the rank of the estimated gradient gradually decreases.
  arXiv  Detail & Related papers  (2024-10-23T13:53:26Z)
• DLO: Dynamic Layer Operation for Efficient Vertical Scaling of LLMs [46.443316184807145]
  We introduce Dynamic Layer Operations (DLO), a novel approach for vertically scaling transformer-based Large Language Models (LLMs) Unlike traditional Mixture-of-Experts (MoE) methods that focus on extending the model width, our approach targets model depth, addressing the redundancy observed across layer representations for various input samples. Experimental results demonstrate that DLO not only outperforms the original unscaled models but also achieves comparable results to densely expanded models with significantly improved efficiency.
  arXiv  Detail & Related papers  (2024-07-03T18:34:08Z)
• Unifying Synergies between Self-supervised Learning and Dynamic Computation [53.66628188936682]
  We present a novel perspective on the interplay between SSL and DC paradigms. We show that it is feasible to simultaneously learn a dense and gated sub-network from scratch in a SSL setting. The co-evolution during pre-training of both dense and gated encoder offers a good accuracy-efficiency trade-off.
  arXiv  Detail & Related papers  (2023-01-22T17:12:58Z)
• Optimization-Derived Learning with Essential Convergence Analysis of Training and Hyper-training [52.39882976848064]
  We design a Generalized Krasnoselskii-Mann (GKM) scheme based on fixed-point iterations as our fundamental ODL module. Under the GKM scheme, a Bilevel Meta Optimization (BMO) algorithmic framework is constructed to solve the optimal training and hyper-training variables together.
  arXiv  Detail & Related papers  (2022-06-16T01:50:25Z)
• Training Generative Adversarial Networks by Solving Ordinary Differential Equations [54.23691425062034]
  We study the continuous-time dynamics induced by GAN training. From this perspective, we hypothesise that instabilities in training GANs arise from the integration error. We experimentally verify that well-known ODE solvers (such as Runge-Kutta) can stabilise training.
  arXiv  Detail & Related papers  (2020-10-28T15:23:49Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)
• Mode-Assisted Unsupervised Learning of Restricted Boltzmann Machines [7.960229223744695]
  We show that properly combining standard gradient updates with an off-gradient direction improves their training dramatically over traditional gradient methods. This approach, which we call mode training, promotes faster training and stability, in addition to lower converged relative entropy (KL divergence) The mode training we suggest is quite versatile, as it can be applied in conjunction with any given gradient method, and is easily extended to more general energy-based neural network structures.
  arXiv  Detail & Related papers  (2020-01-15T21:12:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.