Data-Informed Global Sparseness in Attention Mechanisms for Deep Neural Networks

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Rugina, Ileana, Dangovski, Rumen, Jing, Li, Nakov, Preslav, Soljačić, Marin
Format: Preprint
Veröffentlicht: 2020
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866929346252374016
author Rugina, Ileana
Dangovski, Rumen
Jing, Li
Nakov, Preslav
Soljačić, Marin
author_facet Rugina, Ileana
Dangovski, Rumen
Jing, Li
Nakov, Preslav
Soljačić, Marin
contents Attention mechanisms play a crucial role in the neural revolution of Natural Language Processing (NLP). With the growth of attention-based models, several pruning techniques have been developed to identify and exploit sparseness, making these models more efficient. Most efforts focus on hard-coding attention patterns or pruning attention weights based on training data. We propose Attention Pruning (AP), a framework that observes attention patterns in a fixed dataset and generates a global sparseness mask. AP saves 90% of attention computation for language modeling and about 50% for machine translation and GLUE tasks, maintaining result quality. Our method reveals important distinctions between self- and cross-attention patterns, guiding future NLP research. Our framework can reduce both latency and memory requirements for any attention-based model, aiding in the development of improved models for existing or new NLP applications. We have demonstrated this with encoder and autoregressive transformer models using Triton GPU kernels and make our code publicly available at https://github.com/irugina/AP.
format Preprint
id arxiv_https___arxiv_org_abs_2012_02030
institution arXiv
publishDate 2020
record_format arxiv
spellingShingle Data-Informed Global Sparseness in Attention Mechanisms for Deep Neural Networks
Rugina, Ileana
Dangovski, Rumen
Jing, Li
Nakov, Preslav
Soljačić, Marin
Computation and Language
Attention mechanisms play a crucial role in the neural revolution of Natural Language Processing (NLP). With the growth of attention-based models, several pruning techniques have been developed to identify and exploit sparseness, making these models more efficient. Most efforts focus on hard-coding attention patterns or pruning attention weights based on training data. We propose Attention Pruning (AP), a framework that observes attention patterns in a fixed dataset and generates a global sparseness mask. AP saves 90% of attention computation for language modeling and about 50% for machine translation and GLUE tasks, maintaining result quality. Our method reveals important distinctions between self- and cross-attention patterns, guiding future NLP research. Our framework can reduce both latency and memory requirements for any attention-based model, aiding in the development of improved models for existing or new NLP applications. We have demonstrated this with encoder and autoregressive transformer models using Triton GPU kernels and make our code publicly available at https://github.com/irugina/AP.
title Data-Informed Global Sparseness in Attention Mechanisms for Deep Neural Networks
topic Computation and Language
url https://arxiv.org/abs/2012.02030