Progressive Gradient Flow for Robust N:M Sparsity Training in Transformers
This work addresses a bottleneck in efficient model compression for AI practitioners by improving high-sparsity training, though it is incremental as it builds on existing sparse training methods.
The paper tackles the problem of performance decline in N:M structured sparsity training for Transformers at high sparsity regions (>80%) by identifying elevated gradient noise as a key factor and using decay mechanisms to restrict gradient flow to pruned elements, resulting in up to 2% and 5% improvements in vision and language models, respectively, and up to 2% accuracy gain at iso-training FLOPs.
N:M Structured sparsity has garnered significant interest as a result of relatively modest overhead and improved efficiency. Additionally, this form of sparsity holds considerable appeal for reducing the memory footprint owing to their modest representation overhead. There have been efforts to develop training recipes for N:M structured sparsity, they primarily focus on low-sparsity regions ($\sim$50\%). Nonetheless, performance of models trained using these approaches tends to decline when confronted with high-sparsity regions ($>$80\%). In this work, we study the effectiveness of existing sparse training recipes at \textit{high-sparsity regions} and argue that these methods fail to sustain the model quality on par with low-sparsity regions. We demonstrate that the significant factor contributing to this disparity is the presence of elevated levels of induced noise in the gradient magnitudes. To mitigate this undesirable effect, we employ decay mechanisms to progressively restrict the flow of gradients towards pruned elements. Our approach improves the model quality by up to 2$\%$ and 5$\%$ in vision and language models at high sparsity regime, respectively. We also evaluate the trade-off between model accuracy and training compute cost in terms of FLOPs. At iso-training FLOPs, our method yields better performance compared to conventional sparse training recipes, exhibiting an accuracy improvement of up to 2$\%$. The source code is available at https://github.com/abhibambhaniya/progressive_gradient_flow_nm_sparsity.