Papers › Pay Attention when Required

Pay Attention when Required

9 Sep 2020arXiv:2009.04534archive 2025-07-28

Swetha Mandava, Szymon Migacz, Alex Fit Florea

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.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

Jmkernes/PAR-Transformer-XL mentioned on GitHubtf report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Language ModellingParaphrase IdentificationQuestion AnsweringSentiment Analysis

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Language Modelling Text8 PAR Transformer 24B Bit per Character (BPC) 1.18 #14 of 24 Archive leaderboard report
Language Modelling WikiText-103 PAR Transformer Large Test perplexity 18.4 #35 of 89 Archive leaderboard report
Language Modelling WikiText-103 PAR Transformer Base Test perplexity 22.7 #48 of 89 Archive leaderboard report
Language Modelling enwiki8 PAR Transformer 24B Bit per Character (BPC) 1.11 #1 of 1 Archive leaderboard report
Sentiment Analysis SST-2 Binary classification PAR BERT Base Accuracy 91.6 #50 of 87 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: PAR Transformer

Absolute Position EncodingsAdamAdaptive Input RepresentationsAdaptive SoftmaxAttentionAttention DropoutBERTBPECosine AnnealingDense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerLinear Warmup With Cosine AnnealingLinear Warmup With Linear DecayMulti-Head AttentionPAR TransformerPosition-Wise Feed-Forward LayerReLUResidual ConnectionSoftmaxTransformerTransformer-XLVariational DropoutWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections