A Tensorized Transformer for Language Modeling

TL;DR

This paper introduces a tensorized self-attention mechanism for Transformers that reduces model size and improves performance on language modeling and translation tasks by leveraging tensor decomposition and parameter sharing.

Contribution

It proposes a novel Multi-linear attention model with Block-Term Tensor Decomposition, enhancing efficiency and effectiveness over existing Transformer variants.

Findings

Significant parameter reduction in models.

Performance improvements on language modeling benchmarks.

Outperforms Transformer, Transformer-XL, and tensor train-based models.

Abstract

Latest development of neural models has connected the encoder and decoder through a self-attention mechanism. In particular, Transformer, which is solely based on self-attention, has led to breakthroughs in Natural Language Processing (NLP) tasks. However, the multi-head attention mechanism, as a key component of Transformer, limits the effective deployment of the model to a resource-limited setting. In this paper, based on the ideas of tensor decomposition and parameters sharing, we propose a novel self-attention model (namely Multi-linear attention) with Block-Term Tensor Decomposition (BTD). We test and verify the proposed attention method on three language modeling tasks (i.e., PTB, WikiText-103 and One-billion) and a neural machine translation task (i.e., WMT-2016 English-German). Multi-linear attention can not only largely compress the model parameters but also obtain performance improvements, compared with a number of language modeling approaches, such as Transformer, Transformer-XL, and Transformer with tensor train decomposition.