EcoSpa: Efficient Transformer Training with Coupled Sparsity

TL;DR

EcoSpa is a structured sparse training method for transformers that preserves weight matrix interactions, leading to significant efficiency gains and model compression without specialized hardware.

Contribution

EcoSpa introduces a novel coupled sparsity approach that jointly evaluates and sparsifies weight matrix pairs, maintaining structural relationships for improved performance.

Findings

50% memory reduction in LLaMA-1B training

2.2× model compression on GPT-2-Medium

1.6× inference speedup

Abstract

Transformers have become the backbone of modern AI, yet their high computational demands pose critical system challenges. While sparse training offers efficiency gains, existing methods fail to preserve critical structural relationships between weight matrices that interact multiplicatively in attention and feed-forward layers. This oversight leads to performance degradation at high sparsity levels. We introduce EcoSpa, an efficient structured sparse training method that jointly evaluates and sparsifies coupled weight matrix pairs, preserving their interaction patterns through aligned row/column removal. EcoSpa introduces a new granularity for calibrating structural component importance and performs coupled estimation and sparsification across both pre-training and fine-tuning scenarios. Evaluations demonstrate substantial improvements: EcoSpa enables efficient training of LLaMA-1B with 50\% memory reduction and 21\% faster training, achieves $2.2\times$ model compression on GPT-2-Medium with $2.4$ lower perplexity, and delivers $1.6\times$ inference speedup. The approach uses standard PyTorch operations, requiring no custom hardware or kernels, making efficient transformer training accessible on commodity hardware.