Efficient Low-Rank Hypercomplex Adapter Layers

Adapting large-scale pretrained language models to downstream tasks via fine-tuning is the standard method for achieving state-of-the-art performance on NLP benchmarks.