chunked/nltk_chunking/model_doc_layoutlmv3/chunk_3.txt

Most multimodal pre-trained models use a masked language modeling objective to learn bidirectional representations on the text modality, but they differ in pre-training objectives for the image modality.