Towards Being Parameter-Efficient: A Stratified Sparsely Activated Transformer with Dynamic Capacity
Submission Type: Regular Long Paper
Submission Track: Machine Translation
Submission Track 2: Multilinguality and Linguistic Diversity
Keywords: Stratified Mixture-of-Experts, Parameter-efficiency, Dynamic capacity
TL;DR: We propose Stratified Mixture of Experts (SMoE) models, characterized by a stratified structure that allows for dynamic allocation of capacity to input tokens, taking into consideration the varying capacity requirements of different tokens.
Abstract: Mixture-of-experts (MoE) models that employ sparse activation have demonstrated effectiveness in significantly increasing the number of parameters while maintaining low computational requirements per token. However, recent studies have established that MoE models are inherently parameter-inefficient as the improvement in performance diminishes with an increasing number of experts. We hypothesize this parameter inefficiency is a result of all experts having equal capacity, which may not adequately meet the varying complexity requirements of different tokens or tasks. In light of this, we propose Stratified Mixture of Experts (SMoE) models, which feature a stratified structure and can assign dynamic capacity to different tokens. We demonstrate the effectiveness of SMoE on three multilingual machine translation benchmarks, containing 4, 15, and 94 language pairs, respectively. We show that SMoE outperforms multiple state-of-the-art MoE models with the same or fewer parameters.
Submission Number: 2454
Loading