WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling
This addresses training efficiency and model quality for large language models, particularly in GQA architectures and LoRA fine-tuning, but is incremental as it builds on existing weight optimization concepts.
The paper tackles the lack of systematic optimization of weight patterns during LLM training by proposing WISCA, a weight scaling method that improves convergence quality, showing a 5.6% average improvement on zero-shot validation tasks and 2.12% average reduction in training perplexity.
Transformer architecture gradually dominates the LLM field. Recent advances in training optimization for Transformer-based large language models (LLMs) primarily focus on architectural modifications or optimizer adjustments. However, these approaches lack systematic optimization of weight patterns during training. Weight pattern refers to the distribution and relative magnitudes of weight parameters in a neural network. To address this issue, we propose a Weight Scaling method called WISCA to enhance training efficiency and model quality by strategically improving neural network weight patterns without changing network structures. By rescaling weights while preserving model outputs, WISCA indirectly optimizes the model's training trajectory. Experiments demonstrate that WISCA significantly improves convergence quality (measured by generalization capability and loss reduction), particularly in LLMs with Grouped Query Attention (GQA) architectures and LoRA fine-tuning tasks. Empirical results show 5.6% average improvement on zero-shot validation tasks and 2.12% average reduction in training perplexity across multiple architectures.