English

Differentiable Subset Pruning of Transformer Heads

Computation and Language 2023-07-28 v3

Abstract

Multi-head attention, a collection of several attention mechanisms that independently attend to different parts of the input, is the key ingredient in the Transformer. Recent work has shown, however, that a large proportion of the heads in a Transformer's multi-head attention mechanism can be safely pruned away without significantly harming the performance of the model; such pruning leads to models that are noticeably smaller and faster in practice. Our work introduces a new head pruning technique that we term differentiable subset pruning. Intuitively, our method learns per-head importance variables and then enforces a user-specified hard constraint on the number of unpruned heads. The importance variables are learned via stochastic gradient descent. We conduct experiments on natural language inference and machine translation; we show that differentiable subset pruning performs comparably or better than previous works while offering precise control of the sparsity level.

Keywords

Cite

@article{arxiv.2108.04657,
  title  = {Differentiable Subset Pruning of Transformer Heads},
  author = {Jiaoda Li and Ryan Cotterell and Mrinmaya Sachan},
  journal= {arXiv preprint arXiv:2108.04657},
  year   = {2023}
}

Comments

TACL 2021

R2 v1 2026-06-24T04:59:20.286Z