Linear attention

Linear attention: An attention approximation that reformulates the computation to scale linearly with sequence length rather than quadratically, trading some accuracy for speed.

Browse the full glossary →