Saved in:
Bibliographic Details
Main Authors: Szatkowski, Filip, Wójcik, Bartosz, Piórczyński, Mikołaj, Scardapane, Simone
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.04361
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910694612402176
author Szatkowski, Filip
Wójcik, Bartosz
Piórczyński, Mikołaj
Scardapane, Simone
author_facet Szatkowski, Filip
Wójcik, Bartosz
Piórczyński, Mikołaj
Scardapane, Simone
contents Transformer models can face practical limitations due to their high computational requirements. At the same time, such models exhibit significant activation sparsity, which can be leveraged to reduce the inference cost by converting parts of the network into equivalent Mixture-of-Experts (MoE) layers. Despite the crucial role played by activation sparsity, its impact on this process remains unexplored. We demonstrate that the efficiency of the conversion can be significantly enhanced by a proper regularization of the activation sparsity of the base model. Moreover, motivated by the high variance of the number of activated neurons for different inputs, we introduce a more effective dynamic-$k$ expert selection rule that adjusts the number of executed experts on a per-token basis. To achieve further savings, we extend this approach to multi-head attention projections. Finally, we develop an efficient implementation that translates these computational savings into actual wall-clock speedup. The proposed method, Dense to Dynamic-$k$ Mixture-of-Experts (D2DMoE), outperforms existing approaches on common NLP and vision tasks, reducing inference cost by up to 60% without significantly impacting performance.
format Preprint
id arxiv_https___arxiv_org_abs_2310_04361
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion
Szatkowski, Filip
Wójcik, Bartosz
Piórczyński, Mikołaj
Scardapane, Simone
Machine Learning
Transformer models can face practical limitations due to their high computational requirements. At the same time, such models exhibit significant activation sparsity, which can be leveraged to reduce the inference cost by converting parts of the network into equivalent Mixture-of-Experts (MoE) layers. Despite the crucial role played by activation sparsity, its impact on this process remains unexplored. We demonstrate that the efficiency of the conversion can be significantly enhanced by a proper regularization of the activation sparsity of the base model. Moreover, motivated by the high variance of the number of activated neurons for different inputs, we introduce a more effective dynamic-$k$ expert selection rule that adjusts the number of executed experts on a per-token basis. To achieve further savings, we extend this approach to multi-head attention projections. Finally, we develop an efficient implementation that translates these computational savings into actual wall-clock speedup. The proposed method, Dense to Dynamic-$k$ Mixture-of-Experts (D2DMoE), outperforms existing approaches on common NLP and vision tasks, reducing inference cost by up to 60% without significantly impacting performance.
title Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion
topic Machine Learning
url https://arxiv.org/abs/2310.04361