Pay Attention when Required

Code

• github.com/Jmkernes/PAR-Transformer-XL
  tf★ 7
• github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/LanguageModeling
  tf★ 0

Abstract

Transformer-based models consist of interleaved feed-forward blocks - that capture content meaning, and relatively more expensive self-attention blocks - that capture context meaning. In this paper, we explored trade-offs and ordering of the blocks to improve upon the current Transformer architecture and proposed PAR Transformer. It needs 35% lower compute time than Transformer-XL achieved by replacing ~63% of the self-attention blocks with feed-forward blocks, and retains the perplexity on WikiText-103 language modelling benchmark. We further validated our results on text8 and enwiki8 datasets, as well as on the BERT model.

Tasks

Benchmark Results

Reproductions