Spectrum: Targeted Training on Signal to Noise Ratio
This addresses the problem of high computational costs for researchers and practitioners in AI, though it appears incremental as it builds on existing fine-tuning methods.
The paper tackles the challenge of efficiently post-training large language models by introducing Spectrum, a method that selectively targets layer modules based on signal-to-noise ratio and freezes others, achieving performance comparable to full fine-tuning while reducing GPU memory usage.
Efficiently post-training large language models remains a challenging task due to the vast computational resources required. We present Spectrum, a method that accelerates LLM training by selectively targeting layer modules based on their signal-to-noise ratio (SNR), and freezing the remaining modules. Our approach, which utilizes an algorithm to compute module SNRs prior to training, has shown to effectively match the performance of full fine-tuning while reducing GPU memory usage. Experiments comparing Spectrum to existing methods such as QLoRA demonstrate its effectiveness in terms of model quality and VRAM efficiency in distributed environments.