Sparsity-Accelerated Training for Large Language Models

TL;DR

This paper introduces Sparsity-Accelerated Training (SAT), a method that leverages neuron sparsity in large language models to significantly speed up training without sacrificing performance.

Contribution

The paper proposes a novel sparsity-based training acceleration framework for LLMs, extending neuron importance metrics and introducing a ladder omission rate scheduler.

Findings

45% throughput improvement in continual pre-training

38% training time savings in supervised fine-tuning

Comparable or better performance than standard training

Abstract

Large language models (LLMs) have demonstrated proficiency across various natural language processing (NLP) tasks but often require additional training, such as continual pre-training and supervised fine-tuning. However, the costs associated with this, primarily due to their large parameter count, remain high. This paper proposes leveraging \emph{sparsity} in pre-trained LLMs to expedite this training process. By observing sparsity in activated neurons during forward iterations, we identify the potential for computational speed-ups by excluding inactive neurons. We address associated challenges by extending existing neuron importance evaluation metrics and introducing a ladder omission rate scheduler. Our experiments on Llama-2 demonstrate that Sparsity-Accelerated Training (SAT) achieves comparable or superior performance to standard training while significantly accelerating the process. Specifically, SAT achieves a $45\%$ throughput improvement in continual pre-training and saves $38\%$ training time in supervised fine-tuning in practice. It offers a simple, hardware-agnostic, and easily deployable framework for additional LLM training. Our code is available at https://github.com/OpenDFM/SAT.