chunked/nltk_chunking/model_doc_vit_mae/chunk_18.txt

After pre-training, one "throws away" the decoder used to reconstruct pixels, and one uses the encoder for fine-tuning/linear probing.