chunked/nltk_chunking/model_doc_swiftformer/chunk_10.txt

Using our proposed efficient additive attention, we build a series of models called "SwiftFormer" which achieves state-of-the-art performance in terms of both accuracy and mobile inference speed.