Scale-Distribution Decoupling: Enabling Stable and Effective Training of Large Language Models
This addresses a persistent challenge in pre-training LLMs, offering a practical, lightweight solution for stabilizing training in deep networks.
The paper tackles training instability in large language models (LLMs), particularly Post-Norm Transformers prone to gradient explosion and dissipation, by proposing Scale-Distribution Decoupling (SDD) to decouple weight matrix scale and distribution, resulting in stabilized training across various LLM architectures and outperforming existing techniques.
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.