chunked/nltk_chunking/model_doc_beit/chunk_12.txt

Moreover, large-size BEiT obtains
86.3% only using ImageNet-1K, even outperforming ViT-L with supervised pre-training on ImageNet-22K (85.2%).