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

TL;DR

This paper introduces Attention Pruning (AP), a novel framework that identifies global attention sparseness in NLP models, significantly reducing computation and memory use while maintaining performance.

Contribution

The paper presents a new pruning framework that leverages fixed dataset attention patterns to create global sparsity masks, improving efficiency of attention mechanisms in NLP models.

Findings

AP reduces attention computation by 90% in language modeling.

AP decreases attention computation by about 50% in machine translation and GLUE tasks.

The method reveals key differences between self- and cross-attention patterns.

Abstract

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.