SEER-MoE: Sparse Expert Efficiency through Regularization for Mixture-of-Experts

arXiv:2404.05089v1 Announce Type: cross
Abstract: The advancement of deep learning has led to the emergence of Mixture-of-Experts (MoEs) models, known for their dynamic allocation of computational resources based on input. Despite their promise, MoEs face challenges, particularly in terms of memory requirements. To address this, our work introduces SEER-MoE, a novel two-stage framework for reducing both the memory footprint and compute requirements of pre-trained MoE models. The first stage involves pruning the total number of experts using a heavy-hitters counting …

abstract advancement arxiv challenges computational cs.cl cs.lg deep learning dynamic efficiency emergence expert experts face memory moe novel regularization requirements resources seer stage terms through type work