Transkimmer: Transformer Learns to Layer-wise Skim

TL;DR

Transkimmer introduces a learnable, layer-wise token skimming mechanism in Transformers, significantly reducing computation and speeding up processing while maintaining high accuracy on benchmarks.

Contribution

It proposes an end-to-end trainable skimming method with a parameterized predictor and reparameterization trick for efficient Transformer computation.

Findings

Achieves 10.97x speedup on GLUE benchmark

Maintains less than 1% accuracy degradation

Effectively reduces unnecessary token processing

Abstract

Transformer architecture has become the de-facto model for many machine learning tasks from natural language processing and computer vision. As such, improving its computational efficiency becomes paramount. One of the major computational inefficiency of Transformer-based models is that they spend the identical amount of computation throughout all layers. Prior works have proposed to augment the Transformer model with the capability of skimming tokens to improve its computational efficiency. However, they suffer from not having effectual and end-to-end optimization of the discrete skimming predictor. To address the above limitations, we propose the Transkimmer architecture, which learns to identify hidden state tokens that are not required by each layer. The skimmed tokens are then forwarded directly to the final output, thus reducing the computation of the successive layers. The key idea in Transkimmer is to add a parameterized predictor before each layer that learns to make the skimming decision. We also propose to adopt reparameterization trick and add skim loss for the end-to-end training of Transkimmer. Transkimmer achieves 10.97x average speedup on GLUE benchmark compared with vanilla BERT-base baseline with less than 1% accuracy degradation.