Towards Being Parameter-Efficient: A Stratified Sparsely Activated Transformer with Dynamic Capacity
This addresses the problem of diminishing performance gains with increasing parameters in MoE models for researchers and practitioners in machine translation, representing an incremental improvement.
The paper tackles the parameter inefficiency of mixture-of-experts (MoE) models by proposing Stratified Mixture of Experts (SMoE), which uses a stratified structure to assign dynamic capacity to tokens, and demonstrates its effectiveness by outperforming state-of-the-art MoE models on three multilingual machine translation benchmarks.
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.